diff --git a/.bazelrc b/.bazelrc
new file mode 100644
index 0000000..bdc2c07
--- /dev/null
+++ b/.bazelrc
@@ -0,0 +1,170 @@
+# This file is based on tensorflow's (v2.2.0) .bazelrc found here:
+# https://github.com/tensorflow/tensorflow/blob/v2.2.0/.bazelrc
+
+# Sets the default Apple platform to macOS.
+
+build --apple_platform_type=macos
+
+# Enable using platform specific build settings
+build --enable_platform_specific_config
+
+# Flag to enable remote config. Required starting from TF 2.2.
+common --experimental_repo_remote_exec
+
+# For workaround https://github.com/bazelbuild/bazel/issues/8772 with Bazel >= 0.29.1
+build --java_toolchain=//third_party/toolchains/java:tf_java_toolchain
+build --host_java_toolchain=//third_party/toolchains/java:tf_java_toolchain
+
+# Suppress C++ compiler warnings, otherwise build logs become 10s of MBs.
+build:android --copt=-w
+build:linux --copt=-w
+build:macos --copt=-w
+build:windows --copt=/w
+
+# Android workspace configurations. Should be replaced by an interative configure in the future.
+build --action_env ANDROID_NDK_HOME
+build --action_env ANDROID_NDK_API_LEVEL
+build --action_env ANDROID_BUILD_TOOLS_VERSION
+build --action_env ANDROID_SDK_API_LEVEL
+build --action_env ANDROID_SDK_HOME
+
+# Android configs. Bazel needs to have --cpu and --fat_apk_cpu both set to the
+# target CPU to build transient dependencies correctly. See
+# https://docs.bazel.build/versions/master/user-manual.html#flag--fat_apk_cpu
+
+build:android --crosstool_top=//external:android/crosstool
+build:android --host_crosstool_top=@bazel_tools//tools/cpp:toolchain
+build:android_arm --config=android
+build:android_arm --cpu=armeabi-v7a
+build:android_arm --fat_apk_cpu=armeabi-v7a
+build:android_arm64 --config=android
+build:android_arm64 --cpu=arm64-v8a
+build:android_arm64 --fat_apk_cpu=arm64-v8a
+build:android_x86 --config=android
+build:android_x86 --cpu=x86
+build:android_x86 --fat_apk_cpu=x86
+build:android_x86_64 --config=android
+build:android_x86_64 --cpu=x86_64
+build:android_x86_64 --fat_apk_cpu=x86_64
+
+# iOS configs for each architecture and the fat binary builds.
+build:ios --apple_platform_type=ios
+build:ios --apple_bitcode=embedded --copt=-fembed-bitcode
+build:ios --copt=-Wno-c++11-narrowing
+build:ios_armv7 --config=ios
+build:ios_armv7 --cpu=ios_armv7
+build:ios_arm64 --config=ios
+build:ios_arm64 --cpu=ios_arm64
+build:ios_x86_64 --config=ios
+build:ios_x86_64 --cpu=ios_x86_64
+build:ios_fat --config=ios
+build:ios_fat --ios_multi_cpus=armv7,arm64,x86_64
+
+# By default, build TF in C++ 14 mode.
+build:android --cxxopt=-std=c++14
+build:android --host_cxxopt=-std=c++14
+build:ios --cxxopt=-std=c++14
+build:ios --host_cxxopt=-std=c++14
+build:linux --cxxopt=-std=c++14
+build:linux --host_cxxopt=-std=c++14
+build:macos --cxxopt=-std=c++14
+build:macos --host_cxxopt=-std=c++14
+build:windows --cxxopt=/std:c++14
+build:windows --host_cxxopt=/std:c++14
+
+# Config to use a mostly-static build and disable modular op registration
+# support (this will revert to loading TensorFlow with RTLD_GLOBAL in Python).
+# By default, TensorFlow will build with a dependence on
+# //tensorflow:libtensorflow_framework.so.
+build:monolithic --define framework_shared_object=false
+
+# For projects which use TensorFlow as part of a Bazel build process, putting
+# nothing in a bazelrc will default to a monolithic build. The following line
+# opts in to modular op registration support by default.
+build --define framework_shared_object=true
+
+# ASAN build
+build:asan --strip=never
+build:asan --copt -fsanitize=address
+build:asan --copt -DADDRESS_SANITIZER
+build:asan --copt -O1
+build:asan --copt -g
+build:asan --copt -fno-omit-frame-pointer
+build:asan --linkopt -fsanitize=address
+
+# Flags for open source build, always set to be true.
+build --define open_source_build=true
+test --define open_source_build=true
+
+# dbg config, as a shorthand for '--config=opt -c dbg'
+build:dbg --config=opt -c dbg
+# for now, disable arm_neon. see: https://github.com/tensorflow/tensorflow/issues/33360
+build:dbg --cxxopt -DTF_LITE_DISABLE_X86_NEON
+# AWS SDK must be compiled in release mode. see: https://github.com/tensorflow/tensorflow/issues/37498
+build:dbg --copt -DDEBUG_BUILD
+
+build --define=use_fast_cpp_protos=true
+build --define=allow_oversize_protos=true
+
+build --spawn_strategy=standalone
+build -c opt
+
+# Adding "--cxxopt=-D_GLIBCXX_USE_CXX11_ABI=0" creates parity with TF
+# compilation options. It also addresses memory use due to
+# copy-on-write semantics of std::strings of the older ABI.
+build --cxxopt=-D_GLIBCXX_USE_CXX11_ABI=0
+
+# Make Bazel print out all options from rc files.
+build --announce_rc
+
+# Other build flags.
+build --define=grpc_no_ares=true
+
+# See https://github.com/bazelbuild/bazel/issues/7362 for information on what
+# --incompatible_remove_legacy_whole_archive flag does.
+# This flag is set to true in Bazel 1.0 and newer versions. We tried to migrate
+# Tensorflow to the default, however test coverage wasn't enough to catch the
+# errors.
+# There is ongoing work on Bazel team's side to provide support for transitive
+# shared libraries. As part of migrating to transitive shared libraries, we
+# hope to provide a better mechanism for control over symbol exporting, and
+# then tackle this issue again.
+#
+# TODO: Remove this line once TF doesn't depend on Bazel wrapping all library
+# archives in -whole_archive -no_whole_archive.
+build --noincompatible_remove_legacy_whole_archive
+
+# These are bazel 2.0's incompatible flags. Tensorflow needs to use bazel 2.0.0
+# to use cc_shared_library, as part of the Tensorflow Build Improvements RFC:
+# https://github.com/tensorflow/community/pull/179
+build --noincompatible_prohibit_aapt1
+
+# Build TF with C++ 17 features.
+build:c++17 --cxxopt=-std=c++1z
+build:c++17 --cxxopt=-stdlib=libc++
+build:c++1z --config=c++17
+
+# Enable using platform specific build settings, except when cross-compiling for
+# mobile platforms.
+build --enable_platform_specific_config
+build:android --noenable_platform_specific_config
+build:ios --noenable_platform_specific_config
+
+# Suppress all warning messages.
+build:short_logs --output_filter=DONT_MATCH_ANYTHING
+build:verbose_logs --output_filter=
+build --config=short_logs
+
+# Options to build TensorFlow 1.x or 2.x.
+build:v1 --define=tf_api_version=1
+build:v2 --define=tf_api_version=2
+build:v1 --action_env=TF2_BEHAVIOR=0
+build:v2 --action_env=TF2_BEHAVIOR=1
+build --config=v2
+test --config=v2
+
+# Options from ./configure
+try-import %workspace%/.tf_configure.bazelrc
+
+# Put user-specific options in .bazelrc.user
+try-import %workspace%/.bazelrc.user
diff --git a/BUILD b/BUILD
new file mode 100644
index 0000000..abfcbdd
--- /dev/null
+++ b/BUILD
@@ -0,0 +1,5 @@
+exports_files(
+    [
+        "LICENSE",
+    ],
+)
diff --git a/LICENSE b/LICENSE
new file mode 100644
index 0000000..786bd07
--- /dev/null
+++ b/LICENSE
@@ -0,0 +1,203 @@
+Copyright 2020 The TensorFlow Authors.  All rights reserved.
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
diff --git a/METADATA b/METADATA
new file mode 100644
index 0000000..fb2db71
--- /dev/null
+++ b/METADATA
@@ -0,0 +1,19 @@
+name: "tflite-support"
+description:
+    "TFLite Support is a toolkit that helps users to develop ML and deploy "
+    "TFLite models onto mobile devices. It works cross-Platform and is "
+    "supported on Java, C++ (WIP), and Swift (WIP)."
+
+third_party {
+  url {
+    type: HOMEPAGE
+    value: "https://github.com/tensorflow/tflite-support"
+  }
+  url {
+    type: GIT
+    value: "https://github.com/tensorflow/tflite-support"
+  }
+  version: "v0.1.0"
+  last_upgrade_date { year: 2021 month: 1 day: 14 }
+  license_type: NOTICE
+}
\ No newline at end of file
diff --git a/MODULE_LICENSE_APACHE2 b/MODULE_LICENSE_APACHE2
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/MODULE_LICENSE_APACHE2
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..67d4a8f
--- /dev/null
+++ b/README.md
@@ -0,0 +1,62 @@
+# TensorFlow Lite Support
+
+TFLite Support is a toolkit that helps users to develop ML and deploy TFLite
+models onto mobile devices. It works cross-Platform and is supported on Java,
+C++ (WIP), and Swift (WIP). The TFLite Support project consists of the following
+major components:
+
+*   **TFLite Support Library**: a cross-platform library that helps to
+    deploy TFLite models onto mobile devices.
+*   **TFLite Model Metadata**: (metadata populator and metadata extractor
+    library): includes both human and machine readable information about what a
+    model does and how to use the model.
+*   **TFLite Support Codegen Tool**: an executable that generates model wrapper
+    automatically based on the Support Library and the metadata.
+*   **TFLite Support Task Library**: a flexible and ready-to-use library for
+    common machine learning model types, such as classification and detection,
+    client can also build their own native/Android/iOS inference API on Task
+    Library infra.
+
+TFLite Support library serves different tiers of deployment requirements from
+easy onboarding to fully customizable. There are three major use cases that
+TFLite Support targets at:
+
+*   **Provide ready-to-use APIs for users to interact with the model**. \
+    This is achieved by the TFLite Support Codegen tool, where users can get the
+    model interface (contains ready-to-use APIs) simply by passing the model to
+    the codegen tool. The automatic codegen strategy is designed based on the
+    TFLite metadata.
+
+*   **Provide optimized model interface for popular ML tasks**. \
+    The model interfaces provided by the TFLite Support Task Library are
+    specifically optimized compared to the codegen version in terms of both
+    usability and performance. Users can also swap their own custom models with
+    the default models in each task.
+
+*   **Provide the flexibility to customize model interface and build inference
+    pipelines**. \
+    The TFLite Support Util Library contains varieties of util methods and data
+    structures to perform pre/post processing and data conversion. It is also
+    designed to match the behavior of TensorFlow modules, such as TF.Image and
+    TF.text, ensuring consistency from training to inferencing.
+
+See the
+[documentation on tensorflow.org](https://www.tensorflow.org/lite/inference_with_metadata/overview)
+for more instruction and examples.
+
+## Build Instructions
+
+We use Bazel to build the project. When you're building the Java (Android)
+Utils, you need to set up following env variables correctly:
+
+*   `ANDROID_NDK_HOME`
+*   `ANDROID_SDK_HOME`
+*   `ANDROID_NDK_API_LEVEL`
+*   `ANDROID_SDK_API_LEVEL`
+*   `ANDROID_BUILD_TOOLS_VERSION`
+
+## Contact us
+
+Let us know what you think about TFLite Support by creating a
+[new Github issue](https://github.com/tensorflow/tflite-support/issues/new), or
+email us at tflite-support-team@google.com.
diff --git a/WORKSPACE b/WORKSPACE
new file mode 100644
index 0000000..2194871
--- /dev/null
+++ b/WORKSPACE
@@ -0,0 +1,384 @@
+workspace(name = "org_tensorflow_lite_support")
+
+load("@bazel_tools//tools/build_defs/repo:java.bzl", "java_import_external")
+load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")
+load("@//third_party/py:python_configure.bzl", "python_configure")
+
+http_archive(
+    name = "io_bazel_rules_closure",
+    sha256 = "5b00383d08dd71f28503736db0500b6fb4dda47489ff5fc6bed42557c07c6ba9",
+    strip_prefix = "rules_closure-308b05b2419edb5c8ee0471b67a40403df940149",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/bazelbuild/rules_closure/archive/308b05b2419edb5c8ee0471b67a40403df940149.tar.gz",
+        "https://github.com/bazelbuild/rules_closure/archive/308b05b2419edb5c8ee0471b67a40403df940149.tar.gz",  # 2019-06-13
+    ],
+)
+
+# Apple and Swift rules.
+# https://github.com/bazelbuild/rules_apple/releases
+http_archive(
+    name = "build_bazel_rules_apple",
+    sha256 = "ee9e6073aeb5a65c100cb9c44b0017c937706a4ae03176e14a7e78620a198079",
+    strip_prefix = "rules_apple-5131f3d46794bf227d296c82f30c2499c9de3c5b",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/bazelbuild/rules_apple/archive/5131f3d46794bf227d296c82f30c2499c9de3c5b.tar.gz",
+        "https://github.com/bazelbuild/rules_apple/archive/5131f3d46794bf227d296c82f30c2499c9de3c5b.tar.gz",
+    ],
+)
+
+# https://github.com/bazelbuild/rules_swift/releases
+http_archive(
+    name = "build_bazel_rules_swift",
+    sha256 = "d0833bc6dad817a367936a5f902a0c11318160b5e80a20ece35fb85a5675c886",
+    strip_prefix = "rules_swift-3eeeb53cebda55b349d64c9fc144e18c5f7c0eb8",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/bazelbuild/rules_swift/archive/3eeeb53cebda55b349d64c9fc144e18c5f7c0eb8.tar.gz",
+        "https://github.com/bazelbuild/rules_swift/archive/3eeeb53cebda55b349d64c9fc144e18c5f7c0eb8.tar.gz",
+    ],
+)
+
+# tf-nightly-20200810
+http_archive(
+    name = "org_tensorflow",
+    sha256 = "26c833b7e1873936379e810a39d14700281125257ddda8cd822c89111db6f6ae",
+    strip_prefix = "tensorflow-2.4.0",
+    urls = [
+        "https://github.com/tensorflow/tensorflow/archive/v2.4.0.tar.gz",
+    ],
+    patches = ["@//third_party:tensorflow_lite_ios_build.patch"],
+    patch_args = ["-p1"],
+)
+
+# Set up dependencies. Need to do this before set up TF so that our modification
+# could take effects.
+load("//third_party:repo.bzl", "third_party_http_archive")
+
+# Use our patched gflags which fixes a linking issue.
+load("//third_party/gflags:workspace.bzl", gflags = "repo")
+gflags()
+
+third_party_http_archive(
+    name = "pybind11",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/pybind/pybind11/archive/v2.6.0.tar.gz",
+        "https://github.com/pybind/pybind11/archive/v2.6.0.tar.gz",
+    ],
+    sha256 = "90b705137b69ee3b5fc655eaca66d0dc9862ea1759226f7ccd3098425ae69571",
+    strip_prefix = "pybind11-2.6.0",
+    build_file = "//third_party:pybind11.BUILD",
+)
+
+http_archive(
+    name = "absl_py",
+    sha256 = "603febc9b95a8f2979a7bdb77d2f5e4d9b30d4e0d59579f88eba67d4e4cc5462",
+    strip_prefix = "abseil-py-pypi-v0.9.0",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/abseil/abseil-py/archive/pypi-v0.9.0.tar.gz",
+        "https://github.com/abseil/abseil-py/archive/pypi-v0.9.0.tar.gz",
+    ],
+)
+
+http_archive(
+    name = "six_archive",
+    build_file = "//third_party:six.BUILD",
+    sha256 = "d16a0141ec1a18405cd4ce8b4613101da75da0e9a7aec5bdd4fa804d0e0eba73",
+    strip_prefix = "six-1.12.0",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/pypi.python.org/packages/source/s/six/six-1.12.0.tar.gz",
+        "https://pypi.python.org/packages/source/s/six/six-1.12.0.tar.gz",
+    ],
+)
+
+http_archive(
+    name = "com_google_sentencepiece",
+    strip_prefix = "sentencepiece-1.0.0",
+    sha256 = "c05901f30a1d0ed64cbcf40eba08e48894e1b0e985777217b7c9036cac631346",
+    urls = [
+        "https://github.com/google/sentencepiece/archive/1.0.0.zip",
+    ],
+)
+
+http_archive(
+    name = "org_tensorflow_text",
+    sha256 = "f64647276f7288d1b1fe4c89581d51404d0ce4ae97f2bcc4c19bd667549adca8",
+    strip_prefix = "text-2.2.0",
+    urls = [
+        "https://github.com/tensorflow/text/archive/v2.2.0.zip",
+    ],
+    patches = ["@//third_party:tensorflow_text_remove_tf_deps.patch"],
+    patch_args = ["-p1"],
+    repo_mapping = {"@com_google_re2": "@com_googlesource_code_re2"},
+)
+
+http_archive(
+    name = "com_googlesource_code_re2",
+    sha256 = "d070e2ffc5476c496a6a872a6f246bfddce8e7797d6ba605a7c8d72866743bf9",
+    strip_prefix = "re2-506cfa4bffd060c06ec338ce50ea3468daa6c814",
+    urls = [
+        "https://github.com/google/re2/archive/506cfa4bffd060c06ec338ce50ea3468daa6c814.tar.gz",
+    ],
+)
+
+# ABSL cpp library lts_2020_02_25
+# Needed for absl/status
+http_archive(
+    name = "com_google_absl",
+    build_file = "//third_party:com_google_absl.BUILD",
+    urls = [
+        "https://github.com/abseil/abseil-cpp/archive/20200225.tar.gz",
+    ],
+    # Remove after https://github.com/abseil/abseil-cpp/issues/326 is solved.
+    patches = [
+        "@//third_party:com_google_absl_f863b622fe13612433fdf43f76547d5edda0c93001.diff"
+    ],
+    patch_args = [
+        "-p1",
+    ],
+    strip_prefix = "abseil-cpp-20200225",
+    sha256 = "728a813291bdec2aa46eab8356ace9f75ac2ed9dfe2df5ab603c4e6c09f1c353"
+)
+
+http_archive(
+    name = "com_google_glog",
+    sha256 = "1ee310e5d0a19b9d584a855000434bb724aa744745d5b8ab1855c85bff8a8e21",
+    strip_prefix = "glog-028d37889a1e80e8a07da1b8945ac706259e5fd8",
+    urls = [
+        "https://mirror.bazel.build/github.com/google/glog/archive/028d37889a1e80e8a07da1b8945ac706259e5fd8.tar.gz",
+        "https://github.com/google/glog/archive/028d37889a1e80e8a07da1b8945ac706259e5fd8.tar.gz",
+    ],
+)
+
+
+http_archive(
+    name = "zlib",
+    build_file = "//third_party:zlib.BUILD",
+    sha256 = "c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1",
+    strip_prefix = "zlib-1.2.11",
+    urls = [
+        "http://mirror.bazel.build/zlib.net/fossils/zlib-1.2.11.tar.gz",
+        "http://zlib.net/fossils/zlib-1.2.11.tar.gz",  # 2017-01-15
+    ],
+)
+
+http_archive(
+    name = "org_libzip",
+    build_file = "//third_party:libzip.BUILD",
+    sha256 = "a5d22f0c87a2625450eaa5e10db18b8ee4ef17042102d04c62e311993a2ba363",
+    strip_prefix = "libzip-rel-1-5-1",
+    urls = [
+        # Bazel does not like the official download link at libzip.org,
+        # so use the GitHub release tag.
+        "https://mirror.bazel.build/github.com/nih-at/libzip/archive/rel-1-5-1.zip",
+        "https://github.com/nih-at/libzip/archive/rel-1-5-1.zip",
+    ],
+)
+
+http_archive(
+    name = "libyuv",
+    urls = ["https://chromium.googlesource.com/libyuv/libyuv/+archive/6d603ec3f57dafddc424ef895e5d903915e94ba6.tar.gz"],
+    # Adding the constrain of sha256 and strip_prefix will cause failure.
+    # It seems that the downloaded libyuv was different every time, so that
+    # the specified sha256 and strip_prefix cannot match.
+    # sha256 = "ce196c72858456baa8022fa4a0dc18b77d619265dbc0e3d58e25ad15ca402522",
+    # strip_prefix = "libyuv-6d603ec3f57dafddc424ef895e5d903915e94ba6",
+    build_file = "//third_party:libyuv.BUILD",
+)
+
+http_archive(
+    name = "stblib",
+    strip_prefix = "stb-b42009b3b9d4ca35bc703f5310eedc74f584be58",
+    sha256 = "13a99ad430e930907f5611325ec384168a958bf7610e63e60e2fd8e7b7379610",
+    urls = ["https://github.com/nothings/stb/archive/b42009b3b9d4ca35bc703f5310eedc74f584be58.tar.gz"],
+    build_file = "//third_party:stblib.BUILD",
+)
+
+http_archive(
+    name = "google_toolbox_for_mac",
+    url = "https://github.com/google/google-toolbox-for-mac/archive/v2.2.1.zip",
+    sha256 = "e3ac053813c989a88703556df4dc4466e424e30d32108433ed6beaec76ba4fdc",
+    strip_prefix = "google-toolbox-for-mac-2.2.1",
+    build_file = "@//third_party:google_toolbox_for_mac.BUILD",
+)
+
+http_archive(
+    name = "utf_archive",
+    build_file = "@//third_party:utf.BUILD",
+    sha256 = "262a902f622dcd28e05b8a4be10da0aa3899050d0be8f4a71780eed6b2ea65ca",
+    urls = [
+        "https://mirror.bazel.build/9fans.github.io/plan9port/unix/libutf.tgz",
+        "https://9fans.github.io/plan9port/unix/libutf.tgz",
+    ],
+)
+
+http_archive(
+    name = "icu",
+    strip_prefix = "icu-release-64-2",
+    sha256 = "dfc62618aa4bd3ca14a3df548cd65fe393155edd213e49c39f3a30ccd618fc27",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/unicode-org/icu/archive/release-64-2.zip",
+        "https://github.com/unicode-org/icu/archive/release-64-2.zip",
+    ],
+    build_file = "@//third_party:icu.BUILD",
+)
+
+http_archive(
+    name = "fft2d",
+    build_file = "@//third_party/fft2d:fft2d.BUILD",
+    sha256 = "5f4dabc2ae21e1f537425d58a49cdca1c49ea11db0d6271e2a4b27e9697548eb",
+    strip_prefix = "OouraFFT-1.0",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/petewarden/OouraFFT/archive/v1.0.tar.gz",
+        "https://github.com/petewarden/OouraFFT/archive/v1.0.tar.gz",
+    ],
+)
+
+http_archive(
+    name = "darts_clone",
+    build_file = "@//third_party:darts_clone.BUILD",
+    sha256 = "c97f55d05c98da6fcaf7f9ecc6a6dc6bc5b18b8564465f77abff8879d446491c",
+    strip_prefix = "darts-clone-e40ce4627526985a7767444b6ed6893ab6ff8983",
+    urls = [
+        "https://github.com/s-yata/darts-clone/archive/e40ce4627526985a7767444b6ed6893ab6ff8983.zip",
+    ],
+)
+
+http_archive(
+    name = "com_google_protobuf",
+    sha256 = "a79d19dcdf9139fa4b81206e318e33d245c4c9da1ffed21c87288ed4380426f9",
+    strip_prefix = "protobuf-3.11.4",
+    urls = ["https://github.com/protocolbuffers/protobuf/archive/v3.11.4.tar.gz"],
+    patches = [
+        "@//third_party:com_google_protobuf_fixes.diff"
+    ],
+    patch_args = [
+        "-p1",
+    ],
+)
+
+# AutoValue 1.6+ shades Guava, Auto Common, and JavaPoet. That's OK
+# because none of these jars become runtime dependencies.
+java_import_external(
+    name = "com_google_auto_value",
+    jar_sha256 = "fd811b92bb59ae8a4cf7eb9dedd208300f4ea2b6275d726e4df52d8334aaae9d",
+    jar_urls = [
+        "https://mirror.bazel.build/repo1.maven.org/maven2/com/google/auto/value/auto-value/1.6/auto-value-1.6.jar",
+        "https://repo1.maven.org/maven2/com/google/auto/value/auto-value/1.6/auto-value-1.6.jar",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+    generated_rule_name = "processor",
+    exports = ["@com_google_auto_value_annotations"],
+    extra_build_file_content = "\n".join([
+        "java_plugin(",
+        "    name = \"AutoAnnotationProcessor\",",
+        "    output_licenses = [\"unencumbered\"],",
+        "    processor_class = \"com.google.auto.value.processor.AutoAnnotationProcessor\",",
+        "    tags = [\"annotation=com.google.auto.value.AutoAnnotation;genclass=${package}.AutoAnnotation_${outerclasses}${classname}_${methodname}\"],",
+        "    deps = [\":processor\"],",
+        ")",
+        "",
+        "java_plugin(",
+        "    name = \"AutoOneOfProcessor\",",
+        "    output_licenses = [\"unencumbered\"],",
+        "    processor_class = \"com.google.auto.value.processor.AutoOneOfProcessor\",",
+        "    tags = [\"annotation=com.google.auto.value.AutoValue;genclass=${package}.AutoOneOf_${outerclasses}${classname}\"],",
+        "    deps = [\":processor\"],",
+        ")",
+        "",
+        "java_plugin(",
+        "    name = \"AutoValueProcessor\",",
+        "    output_licenses = [\"unencumbered\"],",
+        "    processor_class = \"com.google.auto.value.processor.AutoValueProcessor\",",
+        "    tags = [\"annotation=com.google.auto.value.AutoValue;genclass=${package}.AutoValue_${outerclasses}${classname}\"],",
+        "    deps = [\":processor\"],",
+        ")",
+        "",
+        "java_library(",
+        "    name = \"com_google_auto_value\",",
+        "    exported_plugins = [",
+        "        \":AutoAnnotationProcessor\",",
+        "        \":AutoOneOfProcessor\",",
+        "        \":AutoValueProcessor\",",
+        "    ],",
+        "    exports = [\"@com_google_auto_value_annotations\"],",
+        ")",
+    ]),
+)
+
+# Auto value annotations
+java_import_external(
+    name = "com_google_auto_value_annotations",
+    jar_sha256 = "d095936c432f2afc671beaab67433e7cef50bba4a861b77b9c46561b801fae69",
+    jar_urls = [
+        "https://mirror.bazel.build/repo1.maven.org/maven2/com/google/auto/value/auto-value-annotations/1.6/auto-value-annotations-1.6.jar",
+        "https://repo1.maven.org/maven2/com/google/auto/value/auto-value-annotations/1.6/auto-value-annotations-1.6.jar",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+    neverlink = True,
+    default_visibility = ["@com_google_auto_value//:__pkg__"],
+)
+
+load("//third_party/flatbuffers:workspace.bzl", flatbuffers = "repo")
+
+flatbuffers()
+# Set up TF.
+load("@org_tensorflow//tensorflow:workspace.bzl", "tf_workspace")
+tf_workspace(tf_repo_name="@org_tensorflow")
+
+load("//third_party/tensorflow:tf_configure.bzl", "tf_configure")
+tf_configure(name = "local_config_tf")
+
+# TF submodule compilation doesn't take care of grpc deps. Do it manually here.
+load("@com_github_grpc_grpc//bazel:grpc_deps.bzl", "grpc_deps")
+grpc_deps()
+
+load(
+    "@build_bazel_rules_apple//apple:repositories.bzl",
+    "apple_rules_dependencies",
+)
+apple_rules_dependencies()
+
+load(
+    "@build_bazel_apple_support//lib:repositories.bzl",
+    "apple_support_dependencies",
+)
+apple_support_dependencies()
+
+load("@upb//bazel:repository_defs.bzl", "bazel_version_repository")
+bazel_version_repository(name = "bazel_version")
+
+
+# Set up Android.
+load("//third_party/android:android_configure.bzl", "android_configure")
+android_configure(name="local_config_android")
+load("@local_config_android//:android.bzl", "android_workspace")
+android_workspace()
+
+python_configure(name = "local_config_python")
+
+
+# Maven dependencies.
+
+RULES_JVM_EXTERNAL_TAG = "3.2"
+
+http_archive(
+    name = "rules_jvm_external",
+    strip_prefix = "rules_jvm_external-%s" % RULES_JVM_EXTERNAL_TAG,
+    sha256 = "82262ff4223c5fda6fb7ff8bd63db8131b51b413d26eb49e3131037e79e324af",
+    url = "https://github.com/bazelbuild/rules_jvm_external/archive/%s.zip" % RULES_JVM_EXTERNAL_TAG,
+)
+
+load("@rules_jvm_external//:defs.bzl", "maven_install")
+
+maven_install(
+    artifacts = [
+        "androidx.annotation:annotation:aar:1.1.0",
+    ],
+    repositories = [
+        "https://jcenter.bintray.com",
+        "https://maven.google.com",
+        "https://dl.google.com/dl/android/maven2",
+        "https://repo1.maven.org/maven2",
+    ],
+    fetch_sources = True,
+    version_conflict_policy = "pinned",
+)
diff --git a/tensorflow_lite_support/BUILD b/tensorflow_lite_support/BUILD
new file mode 100644
index 0000000..f123f1f
--- /dev/null
+++ b/tensorflow_lite_support/BUILD
@@ -0,0 +1,38 @@
+# TFLite Support is a toolkit that helps users to develop ML and deploy TFLite
+# models onto mobile devices.
+
+package(
+    default_visibility = ["//visibility:private"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files(["LICENSE"])
+
+# LINT.IfChange
+package_group(
+    name = "users",
+    packages = [
+        # tensorflow_examples/... dep,
+        "//tensorflow_lite_support/...",
+        "//third_party/tensorflow_models/...",
+    ],
+)
+# Remove internal path from tensorflow_lite_support:users in the copybara file.
+# LINT.ThenChange(//tensorflow_lite_support/copy.bara.sky)
+
+# Config setting for determining if we are building for Android.
+config_setting(
+    name = "android",
+    values = {"crosstool_top": "//external:android/crosstool"},
+    visibility = ["//visibility:public"],
+)
+
+# Config setting for determining if we are building for macos.
+config_setting(
+    name = "macos",
+    values = {
+        "apple_platform_type": "macos",
+        "cpu": "darwin",
+    },
+    visibility = ["//visibility:public"],
+)
diff --git a/tensorflow_lite_support/cc/BUILD b/tensorflow_lite_support/cc/BUILD
new file mode 100644
index 0000000..b19bfde
--- /dev/null
+++ b/tensorflow_lite_support/cc/BUILD
@@ -0,0 +1,25 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "common",
+    srcs = [
+        "common.cc",
+    ],
+    hdrs = ["common.h"],
+    deps = [
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:cord",
+    ],
+)
+
+config_setting(
+    name = "tflite_use_c_api",
+    values = {
+        "copt": "-DTFLITE_USE_C_API",
+    },
+    visibility = ["//tensorflow_lite_support:__subpackages__"],
+)
diff --git a/tensorflow_lite_support/cc/common.cc b/tensorflow_lite_support/cc/common.cc
new file mode 100644
index 0000000..47dd3bc
--- /dev/null
+++ b/tensorflow_lite_support/cc/common.cc
@@ -0,0 +1,35 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/common.h"
+
+#include "absl/strings/cord.h"
+#include "absl/strings/str_cat.h"
+
+namespace tflite {
+namespace support {
+
+absl::Status CreateStatusWithPayload(absl::StatusCode canonical_code,
+                                     absl::string_view message,
+                                     TfLiteSupportStatus tfls_code) {
+  // NOTE: Ignores `message` if the canonical code is ok.
+  absl::Status status = absl::Status(canonical_code, message);
+  // NOTE: Does nothing if the canonical code is ok.
+  status.SetPayload(kTfLiteSupportPayload, absl::Cord(absl::StrCat(tfls_code)));
+  return status;
+}
+
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/common.h b/tensorflow_lite_support/cc/common.h
new file mode 100644
index 0000000..50c8dc4
--- /dev/null
+++ b/tensorflow_lite_support/cc/common.h
@@ -0,0 +1,167 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_COMMON_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_COMMON_H_
+
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+
+namespace tflite {
+namespace support {
+
+// Name (aka type URL key) of the `absl::Status` payload which contains a
+// stringified `TfLiteSupportStatus` code (see below).
+constexpr absl::string_view kTfLiteSupportPayload =
+    "tflite::support::TfLiteSupportStatus";
+
+// Error codes for TensorFlow Lite Support (TFLS) C++ APIs.
+//
+// Such codes capture errors encountered in the TFLS layer. They complement all
+// the other type of errors that occur in the lower-level TF Lite codebase (see
+// `TfLiteStatus` codes).
+//
+// At runtime, such codes are meant to be attached (where applicable) to a
+// `absl::Status` in a key-value manner with `kTfLiteSupportPayload` as key and
+// stringifed error code as value (aka payload). This logic is encapsulated in
+// the `CreateStatusWithPayload` helper below for convenience.
+//
+// The returned status includes:
+// 1. The canonical error code (INVALID_ARGUMENT)
+// 2. The fine-grained error message ("Invalid metadata ...")
+// 3. The specific TFLS code as a payload (kMetadataInvalidSchemaVersionError)
+enum class TfLiteSupportStatus {
+  // Generic error codes.
+
+  // Success.
+  kOk = 0,
+  // Unspecified error.
+  kError = 1,
+  // Invalid argument specified.
+  kInvalidArgumentError = 2,
+  // Invalid FlatBuffer file or buffer specified.
+  kInvalidFlatBufferError = 3,
+
+  // File I/O error codes.
+
+  // No such file.
+  kFileNotFoundError = 100,
+  // Permission issue.
+  kFilePermissionDeniedError,
+  // I/O error when reading file.
+  kFileReadError,
+  // I/O error when mmap-ing file.
+  kFileMmapError,
+
+  // TensorFlow Lite metadata error codes.
+
+  // Unexpected schema version (aka file_identifier) in the Metadata FlatBuffer.
+  kMetadataInvalidSchemaVersionError = 200,
+  // No such associated file within metadata, or file has not been packed.
+  kMetadataAssociatedFileNotFoundError,
+  // ZIP I/O error when unpacking an associated file.
+  kMetadataAssociatedFileZipError,
+  // Inconsistency error between the metadata and actual TF Lite model.
+  // E.g.: number of labels and output tensor values differ.
+  kMetadataInconsistencyError,
+  // Invalid process units specified.
+  // E.g.: multiple ProcessUnits with the same type for a given tensor.
+  kMetadataInvalidProcessUnitsError,
+  // Inconsistency error with the number of labels.
+  // E.g.: label files for different locales have a different number of labels.
+  kMetadataNumLabelsMismatchError,
+  // Score calibration parameters parsing error.
+  // E.g.: too many parameters provided in the corresponding associated file.
+  kMetadataMalformedScoreCalibrationError,
+  // Unexpected number of subgraphs for the current task.
+  // E.g.: image classification expects a single subgraph.
+  kMetadataInvalidNumSubgraphsError,
+  // A given tensor requires NormalizationOptions but none were found.
+  // E.g.: float input tensor requires normalization to preprocess input images.
+  kMetadataMissingNormalizationOptionsError,
+  // Invalid ContentProperties specified.
+  // E.g. expected ImageProperties, got BoundingBoxProperties.
+  kMetadataInvalidContentPropertiesError,
+  // Metadata is mandatory but was not found.
+  // E.g. current task requires TFLite Model Metadata but none was found.
+  kMetadataNotFoundError,
+  // Associated TENSOR_AXIS_LABELS or TENSOR_VALUE_LABELS file is mandatory but
+  // none was found or it was empty.
+  // E.g. current task requires labels but none were found.
+  kMetadataMissingLabelsError,
+  // The ProcessingUnit for tokenizer is not correctly configured.
+  // E.g BertTokenizer doesn't have a valid vocab file associated.
+  kMetadataInvalidTokenizerError,
+
+  // Input tensor(s) error codes.
+
+  // Unexpected number of input tensors for the current task.
+  // E.g. current task expects a single input tensor.
+  kInvalidNumInputTensorsError = 300,
+  // Unexpected input tensor dimensions for the current task.
+  // E.g.: only 4D input tensors supported.
+  kInvalidInputTensorDimensionsError,
+  // Unexpected input tensor type for the current task.
+  // E.g.: current task expects a uint8 pixel image as input.
+  kInvalidInputTensorTypeError,
+  // Unexpected input tensor bytes size.
+  // E.g.: size in bytes does not correspond to the expected number of pixels.
+  kInvalidInputTensorSizeError,
+  // No correct input tensor found for the model.
+  // E.g.: input tensor name is not part of the text model's input tensors.
+  kInputTensorNotFoundError,
+
+  // Output tensor(s) error codes.
+
+  // Unexpected output tensor dimensions for the current task.
+  // E.g.: only a batch size of 1 is supported.
+  kInvalidOutputTensorDimensionsError = 400,
+  // Unexpected input tensor type for the current task.
+  // E.g.: multi-head model with different output tensor types.
+  kInvalidOutputTensorTypeError,
+  // No correct output tensor found for the model.
+  // E.g.: output tensor name is not part of the text model's output tensors.
+  kOutputTensorNotFoundError,
+  // Unexpected number of output tensors for the current task.
+  // E.g.: current task expects a single output tensor.
+  kInvalidNumOutputTensorsError,
+
+  // Image processing error codes.
+
+  // Unspecified image processing failures.
+  kImageProcessingError = 500,
+  // Unexpected input or output buffer metadata.
+  // E.g.: rotate RGBA buffer to Grayscale buffer by 90 degrees.
+  kImageProcessingInvalidArgumentError,
+  // Image processing operation failures.
+  // E.g. libyuv rotation failed for an unknown reason.
+  kImageProcessingBackendError,
+};
+
+// Convenience helper to create an `absl::Status` augmented with the
+// fine-grained `tfls_code` attached as payload under the
+// `kTfLiteSupportPayload` type URL key.
+//
+// This should only be used for non-ok codes since otherwise it does nothing
+// more than returning an object identical to an OK status. See `absl::Status`
+// for more details.
+absl::Status CreateStatusWithPayload(
+    absl::StatusCode canonical_code, absl::string_view message,
+    tflite::support::TfLiteSupportStatus tfls_code =
+        tflite::support::TfLiteSupportStatus::kError);
+
+}  // namespace support
+}  // namespace tflite
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_COMMON_H_
diff --git a/tensorflow_lite_support/cc/port/BUILD b/tensorflow_lite_support/cc/port/BUILD
new file mode 100644
index 0000000..195d5a1
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/BUILD
@@ -0,0 +1,73 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "statusor",
+    hdrs = [
+        "statusor.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/port/default:statusor",
+    ],
+)
+
+cc_library(
+    name = "status_macros",
+    hdrs = [
+        "status_macros.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/port/default:status_macros",
+    ],
+)
+
+cc_library(
+    name = "tflite_wrapper",
+    hdrs = ["tflite_wrapper.h"],
+    deps = ["//tensorflow_lite_support/cc/port/default:tflite_wrapper"],
+)
+
+# This is identical to the rule above, except that it gets built with
+# '-DTFLITE_USE_C_API'. This rule is used for unit tests that verify things
+# work correctly when built with TFLITE_USE_C_API defined.
+cc_library(
+    name = "tflite_wrapper_with_c_api_for_test",
+    testonly = 1,
+    hdrs = ["tflite_wrapper.h"],
+    deps = [
+        "//intelligence/mobile_acceleration/proto:allowlist_portable_proto",
+        "//intelligence/mobile_acceleration/support_library:tflite_wrapper_with_c_api_for_test",
+    ],
+)
+
+cc_library(
+    name = "integral_types",
+    hdrs = ["integral_types.h"],
+)
+
+cc_library(
+    name = "gtest",
+    testonly = 1,
+    hdrs = [
+        "gmock.h",
+        "gtest.h",
+    ],
+    deps = [
+        "//testing/base/public:gunit_for_library_testonly",
+    ],
+)
+
+cc_library(
+    name = "gtest_main",
+    testonly = 1,
+    hdrs = [
+        "benchmark.h",
+        "gmock.h",
+        "gtest.h",
+    ],
+    deps = [
+        "@com_google_googletest//:gtest_main",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/port/benchmark.h b/tensorflow_lite_support/cc/port/benchmark.h
new file mode 100644
index 0000000..74bc1a6
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/benchmark.h
@@ -0,0 +1,21 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_BENCHMARK_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_BENCHMARK_H_
+
+#include "gtest/benchmark.h"
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_BENCHMARK_H_
diff --git a/tensorflow_lite_support/cc/port/build_defs.bzl b/tensorflow_lite_support/cc/port/build_defs.bzl
new file mode 100644
index 0000000..a8053db
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/build_defs.bzl
@@ -0,0 +1,30 @@
+""".bzl file for TFLite Support open source build configs."""
+
+load("@com_google_protobuf//:protobuf.bzl", "cc_proto_library")
+
+def provided_args(**kwargs):
+    """Returns the keyword arguments omitting None arguments."""
+    return {k: v for k, v in kwargs.items() if v != None}
+
+def support_cc_proto_library(name, srcs, visibility = None, deps = [], cc_deps = [], testonly = 0):
+    """Generate cc_proto_library for TFLite Support open source version.
+
+      Args:
+        name: the name of the cc_proto_library.
+        srcs: the .proto files of the cc_proto_library for Bazel use.
+        visibility: visibility of this target.
+        deps: a list of dependency labels for Bazel use; must be cc_proto_library.
+        testonly: test only proto or not.
+    """
+    _ignore = [deps]
+    cc_proto_library(**provided_args(
+        name = name,
+        srcs = srcs,
+        visibility = visibility,
+        deps = cc_deps,
+        testonly = testonly,
+        cc_libs = ["@com_google_protobuf//:protobuf"],
+        protoc = "@com_google_protobuf//:protoc",
+        default_runtime = "@com_google_protobuf//:protobuf",
+        alwayslink = 1,
+    ))
diff --git a/tensorflow_lite_support/cc/port/default/BUILD b/tensorflow_lite_support/cc/port/default/BUILD
new file mode 100644
index 0000000..3f6e9e9
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/default/BUILD
@@ -0,0 +1,50 @@
+package(
+    default_visibility = [
+        "//tensorflow_lite_support/cc/port:__pkg__",
+        "//tensorflow_lite_support/cc/test:__pkg__",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "statusor",
+    srcs = ["statusor.cc"],
+    hdrs = [
+        "statusor.h",
+        "statusor_internals.h",
+    ],
+    deps = [
+        "@com_google_absl//absl/base:core_headers",
+        "@com_google_absl//absl/meta:type_traits",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/types:variant",
+        "@com_google_absl//absl/utility",
+        "@com_google_glog//:glog",
+    ],
+)
+
+cc_library(
+    name = "status_macros",
+    hdrs = [
+        "status_macros.h",
+    ],
+    deps = [
+        "@com_google_absl//absl/base:core_headers",
+        "@com_google_absl//absl/status",
+    ],
+)
+
+cc_library(
+    name = "tflite_wrapper",
+    srcs = ["tflite_wrapper.cc"],
+    hdrs = [
+        "tflite_wrapper.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "@com_google_absl//absl/status",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/experimental/acceleration/configuration:configuration_cc_proto",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/port/default/status_macros.h b/tensorflow_lite_support/cc/port/default/status_macros.h
new file mode 100644
index 0000000..47476c9
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/default/status_macros.h
@@ -0,0 +1,215 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+// This file is forked from absl.
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUS_MACROS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUS_MACROS_H_
+
+#include "absl/base/optimization.h"
+#include "absl/status/status.h"
+
+// Evaluates an expression that produces a `absl::Status`. If the status is not
+// ok, returns it from the current function.
+//
+// For example:
+//   absl::Status MultiStepFunction() {
+//      RETURN_IF_ERROR(Function(args...));
+//      RETURN_IF_ERROR(foo.Method(args...));
+//     return absl::OkStatus();
+//   }
+#define RETURN_IF_ERROR(expr)                                          \
+  STATUS_MACROS_IMPL_ELSE_BLOCKER_                                     \
+  if (::tflite::support::status_macro_internal::StatusAdaptorForMacros \
+          status_macro_internal_adaptor = {(expr)}) {                  \
+  } else /* NOLINT */                                                  \
+    return status_macro_internal_adaptor.Consume()
+
+// Executes an expression `rexpr` that returns a `tflite::support::StatusOr<T>`.
+// On OK, moves its value into the variable defined by `lhs`, otherwise returns
+// from the current function. By default the error status is returned
+// unchanged, but it may be modified by an `error_expression`. If there is an
+// error, `lhs` is not evaluated; thus any side effects that `lhs` may have
+// only occur in the success case.
+//
+// Interface:
+//
+//   ASSIGN_OR_RETURN(lhs, rexpr)
+//   ASSIGN_OR_RETURN(lhs, rexpr, error_expression);
+//
+// WARNING: if lhs is parenthesized, the parentheses are removed. See examples
+// for more details.
+//
+// WARNING: expands into multiple statements; it cannot be used in a single
+// statement (e.g. as the body of an if statement without {})!
+//
+// Example: Declaring and initializing a new variable (ValueType can be anything
+//          that can be initialized with assignment, including references):
+//   ASSIGN_OR_RETURN(ValueType value, MaybeGetValue(arg));
+//
+// Example: Assigning to an existing variable:
+//   ValueType value;
+//   ASSIGN_OR_RETURN(value, MaybeGetValue(arg));
+//
+// Example: Assigning to an expression with side effects:
+//   MyProto data;
+//   ASSIGN_OR_RETURN(*data.mutable_str(), MaybeGetValue(arg));
+//   // No field "str" is added on error.
+//
+// Example: Assigning to a std::unique_ptr.
+//   ASSIGN_OR_RETURN(std::unique_ptr<T> ptr, MaybeGetPtr(arg));
+//
+// Example: Assigning to a map. Because of C preprocessor
+// limitation, the type used in ASSIGN_OR_RETURN cannot contain comma, so
+// wrap lhs in parentheses:
+//   ASSIGN_OR_RETURN((absl::flat_hash_map<Foo, Bar> my_map), GetMap());
+// Or use auto if the type is obvious enough:
+//   ASSIGN_OR_RETURN(const auto& my_map, GetMapRef());
+//
+// Example: Assigning to structured bindings. The same situation with comma as
+// in map, so wrap the statement in parentheses.
+//   ASSIGN_OR_RETURN((const auto& [first, second]), GetPair());
+
+#define ASSIGN_OR_RETURN(...)                                                \
+  STATUS_MACROS_IMPL_GET_VARIADIC_((__VA_ARGS__,                             \
+                                    STATUS_MACROS_IMPL_ASSIGN_OR_RETURN_3_,  \
+                                    STATUS_MACROS_IMPL_ASSIGN_OR_RETURN_2_)) \
+  (__VA_ARGS__)
+
+// =================================================================
+// == Implementation details, do not rely on anything below here. ==
+// =================================================================
+
+// Some builds do not support C++14 fully yet, using C++11 constexpr technique.
+constexpr bool TFLSHasPotentialConditionalOperator(const char* lhs, int index) {
+  return (index == -1
+              ? false
+              : (lhs[index] == '?'
+                     ? true
+                     : TFLSHasPotentialConditionalOperator(lhs, index - 1)));
+}
+
+// MSVC incorrectly expands variadic macros, splice together a macro call to
+// work around the bug.
+#define STATUS_MACROS_IMPL_GET_VARIADIC_HELPER_(_1, _2, _3, NAME, ...) NAME
+#define STATUS_MACROS_IMPL_GET_VARIADIC_(args) \
+  STATUS_MACROS_IMPL_GET_VARIADIC_HELPER_ args
+
+#define STATUS_MACROS_IMPL_ASSIGN_OR_RETURN_2_(lhs, rexpr) \
+  STATUS_MACROS_IMPL_ASSIGN_OR_RETURN_3_(lhs, rexpr, _)
+#define STATUS_MACROS_IMPL_ASSIGN_OR_RETURN_3_(lhs, rexpr, error_expression) \
+  STATUS_MACROS_IMPL_ASSIGN_OR_RETURN_(                                      \
+      STATUS_MACROS_IMPL_CONCAT_(_status_or_value, __LINE__), lhs, rexpr,    \
+      error_expression)
+#define STATUS_MACROS_IMPL_ASSIGN_OR_RETURN_(statusor, lhs, rexpr,        \
+                                             error_expression)            \
+  auto statusor = (rexpr);                                                \
+  if (ABSL_PREDICT_FALSE(!statusor.ok())) {                               \
+    ::absl::Status _(std::move(statusor).status());                       \
+    (void)_; /* error_expression is allowed to not use this variable */   \
+    return (error_expression);                                            \
+  }                                                                       \
+  {                                                                       \
+    static_assert(                                                        \
+        #lhs[0] != '(' || #lhs[sizeof(#lhs) - 2] != ')' ||                \
+            !TFLSHasPotentialConditionalOperator(#lhs, sizeof(#lhs) - 2), \
+        "Identified potential conditional operator, consider not "        \
+        "using ASSIGN_OR_RETURN");                                        \
+  }                                                                       \
+  STATUS_MACROS_IMPL_UNPARENTHESIZE_IF_PARENTHESIZED(lhs) =               \
+      std::move(statusor).value()
+
+// Internal helpers for macro expansion.
+#define STATUS_MACROS_IMPL_EAT(...)
+#define STATUS_MACROS_IMPL_REM(...) __VA_ARGS__
+#define STATUS_MACROS_IMPL_EMPTY()
+
+// Internal helpers for emptyness arguments check.
+#define STATUS_MACROS_IMPL_IS_EMPTY_INNER(...) \
+  STATUS_MACROS_IMPL_IS_EMPTY_INNER_I(__VA_ARGS__, 0, 1)
+#define STATUS_MACROS_IMPL_IS_EMPTY_INNER_I(e0, e1, is_empty, ...) is_empty
+
+#define STATUS_MACROS_IMPL_IS_EMPTY(...) \
+  STATUS_MACROS_IMPL_IS_EMPTY_I(__VA_ARGS__)
+#define STATUS_MACROS_IMPL_IS_EMPTY_I(...) \
+  STATUS_MACROS_IMPL_IS_EMPTY_INNER(_, ##__VA_ARGS__)
+
+// Internal helpers for if statement.
+#define STATUS_MACROS_IMPL_IF_1(_Then, _Else) _Then
+#define STATUS_MACROS_IMPL_IF_0(_Then, _Else) _Else
+#define STATUS_MACROS_IMPL_IF(_Cond, _Then, _Else)          \
+  STATUS_MACROS_IMPL_CONCAT_(STATUS_MACROS_IMPL_IF_, _Cond) \
+  (_Then, _Else)
+
+// Expands to 1 if the input is parenthesized. Otherwise expands to 0.
+#define STATUS_MACROS_IMPL_IS_PARENTHESIZED(...) \
+  STATUS_MACROS_IMPL_IS_EMPTY(STATUS_MACROS_IMPL_EAT __VA_ARGS__)
+
+// If the input is parenthesized, removes the parentheses. Otherwise expands to
+// the input unchanged.
+#define STATUS_MACROS_IMPL_UNPARENTHESIZE_IF_PARENTHESIZED(...)             \
+  STATUS_MACROS_IMPL_IF(STATUS_MACROS_IMPL_IS_PARENTHESIZED(__VA_ARGS__),   \
+                        STATUS_MACROS_IMPL_REM, STATUS_MACROS_IMPL_EMPTY()) \
+  __VA_ARGS__
+
+// Internal helper for concatenating macro values.
+#define STATUS_MACROS_IMPL_CONCAT_INNER_(x, y) x##y
+#define STATUS_MACROS_IMPL_CONCAT_(x, y) STATUS_MACROS_IMPL_CONCAT_INNER_(x, y)
+
+// The GNU compiler emits a warning for code like:
+//
+//   if (foo)
+//     if (bar) { } else baz;
+//
+// because it thinks you might want the else to bind to the first if.  This
+// leads to problems with code like:
+//
+//   if (do_expr)  RETURN_IF_ERROR(expr) << "Some message";
+//
+// The "switch (0) case 0:" idiom is used to suppress this.
+#define STATUS_MACROS_IMPL_ELSE_BLOCKER_ \
+  switch (0)                             \
+  case 0:                                \
+  default:  // NOLINT
+
+namespace tflite {
+namespace support {
+namespace status_macro_internal {
+
+// Provides a conversion to bool so that it can be used inside an if statement
+// that declares a variable.
+class StatusAdaptorForMacros {
+ public:
+  StatusAdaptorForMacros(const ::absl::Status& status)  // NOLINT
+      : status_(status) {}
+
+  StatusAdaptorForMacros(::absl::Status&& status)  // NOLINT
+      : status_(std::move(status)) {}
+
+  StatusAdaptorForMacros(const StatusAdaptorForMacros&) = delete;
+  StatusAdaptorForMacros& operator=(const StatusAdaptorForMacros&) = delete;
+
+  explicit operator bool() const { return ABSL_PREDICT_TRUE(status_.ok()); }
+
+  ::absl::Status&& Consume() { return std::move(status_); }
+
+ private:
+  ::absl::Status status_;
+};
+
+}  // namespace status_macro_internal
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUS_MACROS_H_
diff --git a/tensorflow_lite_support/cc/port/default/statusor.cc b/tensorflow_lite_support/cc/port/default/statusor.cc
new file mode 100644
index 0000000..5cf1196
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/default/statusor.cc
@@ -0,0 +1,65 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+// This file is forked from absl.
+
+#include "tensorflow_lite_support/cc/port/default/statusor.h"
+
+#include <utility>
+
+#include <glog/logging.h>
+#include "absl/strings/str_cat.h"
+
+namespace tflite {
+namespace support {
+
+BadStatusOrAccess::BadStatusOrAccess(absl::Status status)
+    : status_(std::move(status)) {}
+
+BadStatusOrAccess::~BadStatusOrAccess() = default;
+
+const char* BadStatusOrAccess::what() const noexcept {
+  return "Bad StatusOr access";
+}
+
+const absl::Status& BadStatusOrAccess::status() const { return status_; }
+
+namespace internal_statusor {
+
+void Helper::HandleInvalidStatusCtorArg(absl::Status* status) {
+  const char* kMessage =
+      "An OK status is not a valid constructor argument to StatusOr<T>";
+  LOG(DFATAL) << kMessage;
+  // In optimized builds, we will fall back to ::util::error::INTERNAL.
+  *status = absl::InternalError(kMessage);
+}
+
+void Helper::Crash(const absl::Status& status) {
+  LOG(FATAL) << "Attempting to fetch value instead of handling error "
+             << status;
+  _exit(1);
+}
+
+void ThrowBadStatusOrAccess(absl::Status status) {
+#ifdef ABSL_HAVE_EXCEPTIONS
+  throw BadStatusOrAccess(std::move(status));
+#else
+  LOG(FATAL) << "Attempting to fetch value instead of handling error "
+             << status;
+#endif
+}
+
+}  // namespace internal_statusor
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/port/default/statusor.h b/tensorflow_lite_support/cc/port/default/statusor.h
new file mode 100644
index 0000000..4273e1c
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/default/statusor.h
@@ -0,0 +1,574 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+// This file is forked from absl.
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUSOR_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUSOR_H_
+
+#include <exception>
+#include <initializer_list>
+#include <new>
+#include <string>
+#include <type_traits>
+#include <utility>
+
+#include "absl/base/optimization.h"
+#include "absl/meta/type_traits.h"
+#include "absl/status/status.h"
+#include "absl/types/variant.h"
+#include "absl/utility/utility.h"
+#include "tensorflow_lite_support/cc/port/default/statusor_internals.h"
+
+namespace tflite {
+namespace support {
+
+#ifndef SWIG
+class BadStatusOrAccess : public std::exception {
+ public:
+  explicit BadStatusOrAccess(absl::Status status);
+  ~BadStatusOrAccess() override;
+  const char* what() const noexcept override;
+  const absl::Status& status() const;
+
+ private:
+  absl::Status status_;
+};
+#endif  // !SWIG
+
+// Returned StatusOr objects may not be ignored.
+// Note: Disabled for SWIG as it doesn't parse attributes correctly.  Codesearch
+// doesn't handle ifdefs as part of a class definitions (b/6995610), so we use a
+// forward declaration.
+#ifndef SWIG
+template <typename T>
+class ABSL_MUST_USE_RESULT StatusOr;
+#endif
+
+template <typename T>
+class StatusOr : private internal_statusor::StatusOrData<T>,
+                 private internal_statusor::CopyCtorBase<T>,
+                 private internal_statusor::MoveCtorBase<T>,
+                 private internal_statusor::CopyAssignBase<T>,
+                 private internal_statusor::MoveAssignBase<T> {
+  template <typename U>
+  friend class StatusOr;
+
+  typedef internal_statusor::StatusOrData<T> Base;
+
+ public:
+  typedef T value_type;
+
+  // Constructs a new StatusOr with Status::UNKNOWN status.  This is marked
+  // 'explicit' to try to catch cases like 'return {};', where people think
+  // tflite::support::StatusOr<std::vector<int>> will be initialized with an
+  // empty vector, instead of a Status::UNKNOWN status.
+  explicit StatusOr();
+
+  // StatusOr<T> is copy constructible if T is copy constructible.
+  StatusOr(const StatusOr&) = default;
+  // StatusOr<T> is copy assignable if T is copy constructible and copy
+  // assignable.
+  StatusOr& operator=(const StatusOr&) = default;
+
+#ifndef SWIG
+
+  // StatusOr<T> is move constructible if T is move constructible.
+  StatusOr(StatusOr&&) = default;
+  // StatusOr<T> is moveAssignable if T is move constructible and move
+  // assignable.
+  StatusOr& operator=(StatusOr&&) = default;
+
+  // Converting constructors from StatusOr<U>, when T is constructible from U.
+  // To avoid ambiguity, they are disabled if T is also constructible from
+  // StatusOr<U>. Explicit iff the corresponding construction of T from U is
+  // explicit.
+  template <
+      typename U,
+      absl::enable_if_t<
+          absl::conjunction<
+              absl::negation<std::is_same<T, U>>,
+              std::is_constructible<T, const U&>,
+              std::is_convertible<const U&, T>,
+              absl::negation<
+                  internal_statusor::IsConstructibleOrConvertibleFromStatusOr<
+                      T, U>>>::value,
+          int> = 0>
+  StatusOr(const StatusOr<U>& other)  // NOLINT
+      : Base(static_cast<const typename StatusOr<U>::Base&>(other)) {}
+  template <
+      typename U,
+      absl::enable_if_t<
+          absl::conjunction<
+              absl::negation<std::is_same<T, U>>,
+              std::is_constructible<T, const U&>,
+              absl::negation<std::is_convertible<const U&, T>>,
+              absl::negation<
+                  internal_statusor::IsConstructibleOrConvertibleFromStatusOr<
+                      T, U>>>::value,
+          int> = 0>
+  explicit StatusOr(const StatusOr<U>& other)
+      : Base(static_cast<const typename StatusOr<U>::Base&>(other)) {}
+
+  template <
+      typename U,
+      absl::enable_if_t<
+          absl::conjunction<
+              absl::negation<std::is_same<T, U>>, std::is_constructible<T, U&&>,
+              std::is_convertible<U&&, T>,
+              absl::negation<
+                  internal_statusor::IsConstructibleOrConvertibleFromStatusOr<
+                      T, U>>>::value,
+          int> = 0>
+  StatusOr(StatusOr<U>&& other)  // NOLINT
+      : Base(static_cast<typename StatusOr<U>::Base&&>(other)) {}
+  template <
+      typename U,
+      absl::enable_if_t<
+          absl::conjunction<
+              absl::negation<std::is_same<T, U>>, std::is_constructible<T, U&&>,
+              absl::negation<std::is_convertible<U&&, T>>,
+              absl::negation<
+                  internal_statusor::IsConstructibleOrConvertibleFromStatusOr<
+                      T, U>>>::value,
+          int> = 0>
+  explicit StatusOr(StatusOr<U>&& other)
+      : Base(static_cast<typename StatusOr<U>::Base&&>(other)) {}
+
+  // Conversion copy/move assignment operator, T must be constructible and
+  // assignable from U. Only enable if T cannot be directly assigned from
+  // StatusOr<U>.
+  template <
+      typename U,
+      absl::enable_if_t<
+          absl::conjunction<
+              absl::negation<std::is_same<T, U>>,
+              std::is_constructible<T, const U&>,
+              std::is_assignable<T, const U&>,
+              absl::negation<
+                  internal_statusor::
+                      IsConstructibleOrConvertibleOrAssignableFromStatusOr<
+                          T, U>>>::value,
+          int> = 0>
+  StatusOr& operator=(const StatusOr<U>& other) {
+    this->Assign(other);
+    return *this;
+  }
+  template <
+      typename U,
+      absl::enable_if_t<
+          absl::conjunction<
+              absl::negation<std::is_same<T, U>>, std::is_constructible<T, U&&>,
+              std::is_assignable<T, U&&>,
+              absl::negation<
+                  internal_statusor::
+                      IsConstructibleOrConvertibleOrAssignableFromStatusOr<
+                          T, U>>>::value,
+          int> = 0>
+  StatusOr& operator=(StatusOr<U>&& other) {
+    this->Assign(std::move(other));
+    return *this;
+  }
+
+#endif  // SWIG
+
+  // Constructs a new StatusOr with the given value. After calling this
+  // constructor, this->ok() will be true and the contained value may be
+  // retrieved with value(), operator*(), or operator->().
+  //
+  // NOTE: Not explicit - we want to use StatusOr<T> as a return type
+  // so it is convenient and sensible to be able to do 'return T()'
+  // when the return type is StatusOr<T>.
+  //
+  // REQUIRES: T is copy constructible.
+  // TODO(b/113125838): Replace this constructor with a direct-initialization
+  // constructor.
+  StatusOr(const T& value);
+
+  // Constructs a new StatusOr with the given non-ok status. After calling this
+  // constructor, this->ok() will be false and calls to value() will CHECK-fail.
+  //
+  // NOTE: Not explicit - we want to use StatusOr<T> as a return
+  // value, so it is convenient and sensible to be able to do 'return
+  // Status()' when the return type is StatusOr<T>.
+  //
+  // REQUIRES: !status.ok(). This requirement is DCHECKed.
+  // In optimized builds, passing util::OkStatus() here will have the effect
+  // of passing util::error::INTERNAL as a fallback.
+  StatusOr(const absl::Status& status);
+  StatusOr& operator=(const absl::Status& status);
+
+#ifndef SWIG
+  // Perfect-forwarding value assignment operator.
+  // If `*this` contains a `T` value before the call, the contained value is
+  // assigned from `std::forward<U>(v)`; Otherwise, it is directly-initialized
+  // from `std::forward<U>(v)`.
+  // This function does not participate in overload unless:
+  // 1. `std::is_constructible_v<T, U>` is true,
+  // 2. `std::is_assignable_v<T&, U>` is true.
+  // 3. `std::is_same_v<StatusOr<T>, std::remove_cvref_t<U>>` is false.
+  // 4. Assigning `U` to `T` is not ambiguous:
+  //  If `U` is `StatusOr<V>` and `T` is constructible and assignable from
+  //  both `StatusOr<V>` and `V`, the assignment is considered bug-prone and
+  //  ambiguous thus will fail to compile. For example:
+  //    StatusOr<bool> s1 = true;  // s1.ok() && *s1 == true
+  //    StatusOr<bool> s2 = false;  // s2.ok() && *s2 == false
+  //    s1 = s2;  // ambiguous, `s1 = *s2` or `s1 = bool(s2)`?
+  template <
+      typename U = T,
+      typename = typename std::enable_if<absl::conjunction<
+          std::is_constructible<T, U&&>, std::is_assignable<T&, U&&>,
+          internal_statusor::IsForwardingAssignmentValid<T, U&&>>::value>::type>
+  StatusOr& operator=(U&& v) {
+    this->Assign(std::forward<U>(v));
+    return *this;
+  }
+
+  // Similar to the `const T&` overload.
+  //
+  // REQUIRES: T is move constructible.
+  StatusOr(T&& value);
+
+  // RValue versions of the operations declared above.
+  StatusOr(absl::Status&& status);
+  StatusOr& operator=(absl::Status&& status);
+
+  // Constructs the inner value T in-place using the provided args, using the
+  // T(args...) constructor.
+  template <typename... Args>
+  explicit StatusOr(absl::in_place_t, Args&&... args);
+  template <typename U, typename... Args>
+  explicit StatusOr(absl::in_place_t, std::initializer_list<U> ilist,
+                    Args&&... args);
+
+  // Constructs the inner value T in-place using the provided args, using the
+  // T(U) (direct-initialization) constructor. Only valid if T can be
+  // constructed from a U. Can accept move or copy constructors. Explicit if
+  // U is not convertible to T. To avoid ambiguity, this is disabled if U is
+  // a StatusOr<J>, where J is convertible to T.
+  // Style waiver for implicit conversion granted in cl/209187539.
+  template <typename U = T,
+            absl::enable_if_t<
+                absl::conjunction<
+                    internal_statusor::IsDirectInitializationValid<T, U&&>,
+                    std::is_constructible<T, U&&>,
+                    std::is_convertible<U&&, T>>::value,
+                int> = 0>
+  StatusOr(U&& u)  // NOLINT
+      : StatusOr(absl::in_place, std::forward<U>(u)) {}
+
+  template <typename U = T,
+            absl::enable_if_t<
+                absl::conjunction<
+                    internal_statusor::IsDirectInitializationValid<T, U&&>,
+                    std::is_constructible<T, U&&>,
+                    absl::negation<std::is_convertible<U&&, T>>>::value,
+                int> = 0>
+  explicit StatusOr(U&& u)  // NOLINT
+      : StatusOr(absl::in_place, std::forward<U>(u)) {}
+
+#endif  // SWIG
+
+  // Returns this->status().ok()
+  ABSL_MUST_USE_RESULT bool ok() const { return this->status_.ok(); }
+
+  // Returns a reference to our status. If this contains a T, then
+  // returns util::OkStatus().
+#ifdef SWIG
+  const ::util::Status& status() const;
+#else  // SWIG
+  const absl::Status& status() const&;
+  absl::Status status() &&;
+#endif  // SWIG
+
+  // Returns a reference to the held value if `this->ok()`. Otherwise, throws
+  // `absl::BadStatusOrAccess` if exception is enabled, or `LOG(FATAL)` if
+  // exception is disabled.
+  // If you have already checked the status using `this->ok()` or
+  // `operator bool()`, you probably want to use `operator*()` or `operator->()`
+  // to access the value instead of `value`.
+  // Note: for value types that are cheap to copy, prefer simple code:
+  //
+  //   T value = statusor.value();
+  //
+  // Otherwise, if the value type is expensive to copy, but can be left
+  // in the StatusOr, simply assign to a reference:
+  //
+  //   T& value = statusor.value();  // or `const T&`
+  //
+  // Otherwise, if the value type supports an efficient move, it can be
+  // used as follows:
+  //
+  //   T value = std::move(statusor).value();
+  //
+  // The `std::move` on statusor instead of on the whole expression enables
+  // warnings about possible uses of the statusor object after the move.
+#ifdef SWIG
+  const T& value() const;
+#else  // SWIG
+  const T& value() const&;
+  T& value() &;
+  const T&& value() const&&;
+  T&& value() &&;
+#endif  // SWIG
+
+#ifndef SWIG
+  // Returns a reference to the current value.
+  //
+  // REQUIRES: this->ok() == true, otherwise the behavior is undefined.
+  //
+  // Use this->ok() or `operator bool()` to verify that there is a current
+  // value. Alternatively, see value() for a similar API that guarantees
+  // CHECK-failing if there is no current value.
+  const T& operator*() const&;
+  T& operator*() &;
+  const T&& operator*() const&&;
+  T&& operator*() &&;
+#endif  // SWIG
+
+#ifndef SWIG
+  // Returns a pointer to the current value.
+  //
+  // REQUIRES: this->ok() == true, otherwise the behavior is undefined.
+  //
+  // Use this->ok() or `operator bool()` to verify that there is a current
+  // value.
+  const T* operator->() const;
+  T* operator->();
+#endif  // SWIG
+
+#ifndef SWIG
+  // Returns a copy of the current value if this->ok() == true. Otherwise
+  // returns a default value.
+  template <typename U>
+  T value_or(U&& default_value) const&;
+  template <typename U>
+  T value_or(U&& default_value) &&;
+#endif  // SWIG
+
+  // Ignores any errors. This method does nothing except potentially suppress
+  // complaints from any tools that are checking that errors are not dropped on
+  // the floor.
+  void IgnoreError() const;
+
+#ifndef SWIG
+  // Reconstructs the inner value T in-place using the provided args, using the
+  // T(args...) constructor. Returns reference to the reconstructed `T`.
+  template <typename... Args>
+  T& emplace(Args&&... args) {
+    if (ok()) {
+      this->Clear();
+      this->MakeValue(std::forward<Args>(args)...);
+    } else {
+      this->MakeValue(std::forward<Args>(args)...);
+      this->status_ = absl::OkStatus();
+    }
+    return this->data_;
+  }
+
+  template <
+      typename U, typename... Args,
+      absl::enable_if_t<
+          std::is_constructible<T, std::initializer_list<U>&, Args&&...>::value,
+          int> = 0>
+  T& emplace(std::initializer_list<U> ilist, Args&&... args) {
+    if (ok()) {
+      this->Clear();
+      this->MakeValue(ilist, std::forward<Args>(args)...);
+    } else {
+      this->MakeValue(ilist, std::forward<Args>(args)...);
+      this->status_ = absl::OkStatus();
+    }
+    return this->data_;
+  }
+#endif  // SWIG
+
+ private:
+#ifndef SWIG
+  using internal_statusor::StatusOrData<T>::Assign;
+  template <typename U>
+  void Assign(const StatusOr<U>& other);
+  template <typename U>
+  void Assign(StatusOr<U>&& other);
+#endif  // SWIG
+};
+
+#ifndef SWIG
+////////////////////////////////////////////////////////////////////////////////
+// Implementation details for StatusOr<T>
+
+template <typename T>
+tflite::support::StatusOr<T>::StatusOr()
+    : Base(absl::Status(absl::StatusCode::kUnknown, "")) {}
+
+template <typename T>
+tflite::support::StatusOr<T>::StatusOr(const T& value) : Base(value) {}
+
+template <typename T>
+tflite::support::StatusOr<T>::StatusOr(const absl::Status& status)
+    : Base(status) {}
+
+template <typename T>
+tflite::support::StatusOr<T>& StatusOr<T>::operator=(
+    const absl::Status& status) {
+  this->Assign(status);
+  return *this;
+}
+
+template <typename T>
+tflite::support::StatusOr<T>::StatusOr(T&& value) : Base(std::move(value)) {}
+
+template <typename T>
+tflite::support::StatusOr<T>::StatusOr(absl::Status&& status)
+    : Base(std::move(status)) {}
+
+template <typename T>
+tflite::support::StatusOr<T>& StatusOr<T>::operator=(absl::Status&& status) {
+  this->Assign(std::move(status));
+  return *this;
+}
+
+template <typename T>
+template <typename U>
+inline void StatusOr<T>::Assign(const StatusOr<U>& other) {
+  if (other.ok()) {
+    this->Assign(other.value());
+  } else {
+    this->Assign(other.status());
+  }
+}
+
+template <typename T>
+template <typename U>
+inline void StatusOr<T>::Assign(StatusOr<U>&& other) {
+  if (other.ok()) {
+    this->Assign(std::move(other).value());
+  } else {
+    this->Assign(std::move(other).status());
+  }
+}
+template <typename T>
+template <typename... Args>
+tflite::support::StatusOr<T>::StatusOr(absl::in_place_t, Args&&... args)
+    : Base(absl::in_place, std::forward<Args>(args)...) {}
+
+template <typename T>
+template <typename U, typename... Args>
+tflite::support::StatusOr<T>::StatusOr(absl::in_place_t,
+                                       std::initializer_list<U> ilist,
+                                       Args&&... args)
+    : Base(absl::in_place, ilist, std::forward<Args>(args)...) {}
+
+template <typename T>
+const absl::Status& StatusOr<T>::status() const& {
+  return this->status_;
+}
+template <typename T>
+absl::Status StatusOr<T>::status() && {
+  return ok() ? absl::OkStatus() : std::move(this->status_);
+}
+
+template <typename T>
+const T& StatusOr<T>::value() const& {
+  if (!this->ok()) internal_statusor::ThrowBadStatusOrAccess(this->status_);
+  return this->data_;
+}
+
+template <typename T>
+T& StatusOr<T>::value() & {
+  if (!this->ok()) internal_statusor::ThrowBadStatusOrAccess(this->status_);
+  return this->data_;
+}
+
+template <typename T>
+const T&& StatusOr<T>::value() const&& {
+  if (!this->ok()) {
+    internal_statusor::ThrowBadStatusOrAccess(std::move(this->status_));
+  }
+  return std::move(this->data_);
+}
+
+template <typename T>
+T&& StatusOr<T>::value() && {
+  if (!this->ok()) {
+    internal_statusor::ThrowBadStatusOrAccess(std::move(this->status_));
+  }
+  return std::move(this->data_);
+}
+
+template <typename T>
+const T& StatusOr<T>::operator*() const& {
+  this->EnsureOk();
+  return this->data_;
+}
+
+template <typename T>
+T& StatusOr<T>::operator*() & {
+  this->EnsureOk();
+  return this->data_;
+}
+
+template <typename T>
+const T&& StatusOr<T>::operator*() const&& {
+  this->EnsureOk();
+  return std::move(this->data_);
+}
+
+template <typename T>
+T&& StatusOr<T>::operator*() && {
+  this->EnsureOk();
+  return std::move(this->data_);
+}
+
+template <typename T>
+const T* StatusOr<T>::operator->() const {
+  this->EnsureOk();
+  return &this->data_;
+}
+
+template <typename T>
+T* StatusOr<T>::operator->() {
+  this->EnsureOk();
+  return &this->data_;
+}
+
+template <typename T>
+template <typename U>
+T StatusOr<T>::value_or(U&& default_value) const& {
+  if (ok()) {
+    return this->data_;
+  }
+  return std::forward<U>(default_value);
+}
+
+template <typename T>
+template <typename U>
+T StatusOr<T>::value_or(U&& default_value) && {
+  if (ok()) {
+    return std::move(this->data_);
+  }
+  return std::forward<U>(default_value);
+}
+
+template <typename T>
+void StatusOr<T>::IgnoreError() const {
+  // no-op
+}
+
+#endif  // SWIG
+
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUSOR_H_
diff --git a/tensorflow_lite_support/cc/port/default/statusor_internals.h b/tensorflow_lite_support/cc/port/default/statusor_internals.h
new file mode 100644
index 0000000..56d4661
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/default/statusor_internals.h
@@ -0,0 +1,409 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+// This file is forked from absl.
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUSOR_INTERNALS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUSOR_INTERNALS_H_
+
+#include <type_traits>
+#include <utility>
+
+#include "absl/meta/type_traits.h"
+#include "absl/status/status.h"
+#include "absl/utility/utility.h"
+
+namespace tflite {
+namespace support {
+
+template <typename T>
+class ABSL_MUST_USE_RESULT StatusOr;
+
+namespace internal_statusor {
+
+// Detects whether `T` is constructible or convertible from `StatusOr<U>`.
+template <typename T, typename U>
+using IsConstructibleOrConvertibleFromStatusOr =
+    absl::disjunction<std::is_constructible<T, StatusOr<U>&>,
+                      std::is_constructible<T, const StatusOr<U>&>,
+                      std::is_constructible<T, StatusOr<U>&&>,
+                      std::is_constructible<T, const StatusOr<U>&&>,
+                      std::is_convertible<StatusOr<U>&, T>,
+                      std::is_convertible<const StatusOr<U>&, T>,
+                      std::is_convertible<StatusOr<U>&&, T>,
+                      std::is_convertible<const StatusOr<U>&&, T>>;
+
+// Detects whether `T` is constructible or convertible or assignable from
+// `StatusOr<U>`.
+template <typename T, typename U>
+using IsConstructibleOrConvertibleOrAssignableFromStatusOr =
+    absl::disjunction<IsConstructibleOrConvertibleFromStatusOr<T, U>,
+                      std::is_assignable<T&, StatusOr<U>&>,
+                      std::is_assignable<T&, const StatusOr<U>&>,
+                      std::is_assignable<T&, StatusOr<U>&&>,
+                      std::is_assignable<T&, const StatusOr<U>&&>>;
+
+// Detects whether direct initializing `StatusOr<T>` from `U` is ambiguous, i.e.
+// when `U` is `StatusOr<V>` and `T` is constructible or convertible from `V`.
+template <typename T, typename U>
+struct IsDirectInitializationAmbiguous
+    : public absl::conditional_t<
+          std::is_same<absl::remove_cv_t<absl::remove_reference_t<U>>,
+                       U>::value,
+          std::false_type,
+          IsDirectInitializationAmbiguous<
+              T, absl::remove_cv_t<absl::remove_reference_t<U>>>> {};
+
+template <typename T, typename V>
+struct IsDirectInitializationAmbiguous<T, tflite::support::StatusOr<V>>
+    : public IsConstructibleOrConvertibleFromStatusOr<T, V> {};
+
+// Checks against the constraints of the direction initialization, i.e. when
+// `StatusOr<T>::StatusOr(U&&)` should participate in overload resolution.
+template <typename T, typename U>
+using IsDirectInitializationValid = absl::disjunction<
+    // Short circuits if T is basically U.
+    std::is_same<T, absl::remove_cv_t<absl::remove_reference_t<U>>>,
+    absl::negation<absl::disjunction<
+        std::is_same<tflite::support::StatusOr<T>,
+                     absl::remove_cv_t<absl::remove_reference_t<U>>>,
+        std::is_same<absl::Status,
+                     absl::remove_cv_t<absl::remove_reference_t<U>>>,
+        std::is_same<absl::in_place_t,
+                     absl::remove_cv_t<absl::remove_reference_t<U>>>,
+        IsDirectInitializationAmbiguous<T, U>>>>;
+
+// This trait detects whether `StatusOr<T>::operator=(U&&)` is ambiguous, which
+// is equivalent to whether all the following conditions are met:
+// 1. `U` is `StatusOr<V>`.
+// 2. `T` is constructible and assignable from `V`.
+// 3. `T` is constructible and assignable from `U` (i.e. `StatusOr<V>`).
+// For example, the following code is considered ambiguous:
+// (`T` is `bool`, `U` is `StatusOr<bool>`, `V` is `bool`)
+//   StatusOr<bool> s1 = true;  // s1.ok() && s1.ValueOrDie() == true
+//   StatusOr<bool> s2 = false;  // s2.ok() && s2.ValueOrDie() == false
+//   s1 = s2;  // ambiguous, `s1 = s2.ValueOrDie()` or `s1 = bool(s2)`?
+template <typename T, typename U>
+struct IsForwardingAssignmentAmbiguous
+    : public absl::conditional_t<
+          std::is_same<absl::remove_cv_t<absl::remove_reference_t<U>>,
+                       U>::value,
+          std::false_type,
+          IsForwardingAssignmentAmbiguous<
+              T, absl::remove_cv_t<absl::remove_reference_t<U>>>> {};
+
+template <typename T, typename U>
+struct IsForwardingAssignmentAmbiguous<T, tflite::support::StatusOr<U>>
+    : public IsConstructibleOrConvertibleOrAssignableFromStatusOr<T, U> {};
+
+// Checks against the constraints of the forwarding assignment, i.e. whether
+// `StatusOr<T>::operator(U&&)` should participate in overload resolution.
+template <typename T, typename U>
+using IsForwardingAssignmentValid = absl::disjunction<
+    // Short circuits if T is basically U.
+    std::is_same<T, absl::remove_cv_t<absl::remove_reference_t<U>>>,
+    absl::negation<absl::disjunction<
+        std::is_same<tflite::support::StatusOr<T>,
+                     absl::remove_cv_t<absl::remove_reference_t<U>>>,
+        std::is_same<absl::Status,
+                     absl::remove_cv_t<absl::remove_reference_t<U>>>,
+        std::is_same<absl::in_place_t,
+                     absl::remove_cv_t<absl::remove_reference_t<U>>>,
+        IsForwardingAssignmentAmbiguous<T, U>>>>;
+
+class Helper {
+ public:
+  // Move type-agnostic error handling to the .cc.
+  static void HandleInvalidStatusCtorArg(absl::Status*);
+  ABSL_ATTRIBUTE_NORETURN static void Crash(const absl::Status& status);
+};
+
+// Construct an instance of T in `p` through placement new, passing Args... to
+// the constructor.
+// This abstraction is here mostly for the gcc performance fix.
+template <typename T, typename... Args>
+void PlacementNew(void* p, Args&&... args) {
+#if defined(__GNUC__) && !defined(__clang__)
+  // Teach gcc that 'p' cannot be null, fixing code size issues.
+  if (p == nullptr) __builtin_unreachable();
+#endif
+  new (p) T(std::forward<Args>(args)...);
+}
+
+// Helper base class to hold the data and all operations.
+// We move all this to a base class to allow mixing with the appropriate
+// TraitsBase specialization.
+template <typename T>
+class StatusOrData {
+  template <typename U>
+  friend class StatusOrData;
+
+ public:
+  StatusOrData() = delete;
+
+  StatusOrData(const StatusOrData& other) {
+    if (other.ok()) {
+      MakeValue(other.data_);
+      MakeStatus();
+    } else {
+      MakeStatus(other.status_);
+    }
+  }
+
+  StatusOrData(StatusOrData&& other) noexcept {
+    if (other.ok()) {
+      MakeValue(std::move(other.data_));
+      MakeStatus();
+    } else {
+      MakeStatus(std::move(other.status_));
+    }
+  }
+
+  template <typename U>
+  explicit StatusOrData(const StatusOrData<U>& other) {
+    if (other.ok()) {
+      MakeValue(other.data_);
+      MakeStatus();
+    } else {
+      MakeStatus(other.status_);
+    }
+  }
+
+  template <typename U>
+  explicit StatusOrData(StatusOrData<U>&& other) {
+    if (other.ok()) {
+      MakeValue(std::move(other.data_));
+      MakeStatus();
+    } else {
+      MakeStatus(std::move(other.status_));
+    }
+  }
+
+  template <typename... Args>
+  explicit StatusOrData(absl::in_place_t, Args&&... args)
+      : data_(std::forward<Args>(args)...) {
+    MakeStatus();
+  }
+
+  explicit StatusOrData(const T& value) : data_(value) { MakeStatus(); }
+  explicit StatusOrData(T&& value) : data_(std::move(value)) { MakeStatus(); }
+
+  explicit StatusOrData(const absl::Status& status) : status_(status) {
+    EnsureNotOk();
+  }
+  explicit StatusOrData(absl::Status&& status) : status_(std::move(status)) {
+    EnsureNotOk();
+  }
+
+  StatusOrData& operator=(const StatusOrData& other) {
+    if (this == &other) return *this;
+    if (other.ok())
+      Assign(other.data_);
+    else
+      Assign(other.status_);
+    return *this;
+  }
+
+  StatusOrData& operator=(StatusOrData&& other) {
+    if (this == &other) return *this;
+    if (other.ok())
+      Assign(std::move(other.data_));
+    else
+      Assign(std::move(other.status_));
+    return *this;
+  }
+
+  ~StatusOrData() {
+    if (ok()) {
+      status_.~Status();
+      data_.~T();
+    } else {
+      status_.~Status();
+    }
+  }
+
+  // TODO(b/140189837): Remove the SFINAE condition after cleanup.
+  template <typename U,
+            absl::enable_if_t<std::is_assignable<T&, U&&>::value, int> = 0>
+  void Assign(U&& value) {
+    if (ok()) {
+      data_ = std::forward<U>(value);
+    } else {
+      MakeValue(std::forward<U>(value));
+      status_ = absl::OkStatus();
+    }
+  }
+
+  // TODO(b/140189837): Remove this after cleanup.
+  // This overload is to handle the case where `T` is a `const` type.
+  // `StatusOr` supports assignment for `const` types though it's forbidden by
+  // other standard types like `std::optional`.
+  template <typename U,
+            absl::enable_if_t<!std::is_assignable<T&, U&&>::value, int> = 0>
+  void Assign(U&& value) {
+    if (ok()) {
+      data_.~T();
+      MakeValue(std::forward<U>(value));
+    } else {
+      MakeValue(std::forward<U>(value));
+      status_ = absl::OkStatus();
+    }
+  }
+
+  void Assign(const absl::Status& status) {
+    Clear();
+    status_ = status;
+    EnsureNotOk();
+  }
+
+  void Assign(absl::Status&& status) {
+    Clear();
+    status_ = std::move(status);
+    EnsureNotOk();
+  }
+
+  bool ok() const { return status_.ok(); }
+
+ protected:
+  // status_ will always be active after the constructor.
+  // We make it a union to be able to initialize exactly how we need without
+  // waste.
+  // Eg. in the copy constructor we use the default constructor of Status in
+  // the ok() path to avoid an extra Ref call.
+  union {
+    absl::Status status_;
+  };
+
+  // data_ is active iff status_.ok()==true
+  struct Dummy {};
+  union {
+    // When T is const, we need some non-const object we can cast to void* for
+    // the placement new. dummy_ is that object.
+    Dummy dummy_;
+    T data_;
+  };
+
+  void Clear() {
+    if (ok()) data_.~T();
+  }
+
+  void EnsureOk() const {
+    if (ABSL_PREDICT_FALSE(!ok())) Helper::Crash(status_);
+  }
+
+  void EnsureNotOk() {
+    if (ABSL_PREDICT_FALSE(ok())) Helper::HandleInvalidStatusCtorArg(&status_);
+  }
+
+  // Construct the value (ie. data_) through placement new with the passed
+  // argument.
+  template <typename... Arg>
+  void MakeValue(Arg&&... arg) {
+    internal_statusor::PlacementNew<T>(&dummy_, std::forward<Arg>(arg)...);
+  }
+
+  // Construct the status (ie. status_) through placement new with the passed
+  // argument.
+  template <typename... Args>
+  void MakeStatus(Args&&... args) {
+    internal_statusor::PlacementNew<absl::Status>(&status_,
+                                                  std::forward<Args>(args)...);
+  }
+};
+
+// Helper base classes to allow implicitly deleted constructors and assignment
+// operators in `StatusOr`. For example, `CopyCtorBase` will explicitly delete
+// the copy constructor when T is not copy constructible and `StatusOr` will
+// inherit that behavior implicitly.
+template <typename T, bool = std::is_copy_constructible<T>::value>
+struct CopyCtorBase {
+  CopyCtorBase() = default;
+  CopyCtorBase(const CopyCtorBase&) = default;
+  CopyCtorBase(CopyCtorBase&&) = default;
+  CopyCtorBase& operator=(const CopyCtorBase&) = default;
+  CopyCtorBase& operator=(CopyCtorBase&&) = default;
+};
+
+template <typename T>
+struct CopyCtorBase<T, false> {
+  CopyCtorBase() = default;
+  CopyCtorBase(const CopyCtorBase&) = delete;
+  CopyCtorBase(CopyCtorBase&&) = default;
+  CopyCtorBase& operator=(const CopyCtorBase&) = default;
+  CopyCtorBase& operator=(CopyCtorBase&&) = default;
+};
+
+template <typename T, bool = std::is_move_constructible<T>::value>
+struct MoveCtorBase {
+  MoveCtorBase() = default;
+  MoveCtorBase(const MoveCtorBase&) = default;
+  MoveCtorBase(MoveCtorBase&&) = default;
+  MoveCtorBase& operator=(const MoveCtorBase&) = default;
+  MoveCtorBase& operator=(MoveCtorBase&&) = default;
+};
+
+template <typename T>
+struct MoveCtorBase<T, false> {
+  MoveCtorBase() = default;
+  MoveCtorBase(const MoveCtorBase&) = default;
+  MoveCtorBase(MoveCtorBase&&) = delete;
+  MoveCtorBase& operator=(const MoveCtorBase&) = default;
+  MoveCtorBase& operator=(MoveCtorBase&&) = default;
+};
+
+template <typename T, bool = std::is_copy_constructible<T>::value&&
+                          std::is_copy_assignable<T>::value>
+struct CopyAssignBase {
+  CopyAssignBase() = default;
+  CopyAssignBase(const CopyAssignBase&) = default;
+  CopyAssignBase(CopyAssignBase&&) = default;
+  CopyAssignBase& operator=(const CopyAssignBase&) = default;
+  CopyAssignBase& operator=(CopyAssignBase&&) = default;
+};
+
+template <typename T>
+struct CopyAssignBase<T, false> {
+  CopyAssignBase() = default;
+  CopyAssignBase(const CopyAssignBase&) = default;
+  CopyAssignBase(CopyAssignBase&&) = default;
+  CopyAssignBase& operator=(const CopyAssignBase&) = delete;
+  CopyAssignBase& operator=(CopyAssignBase&&) = default;
+};
+
+template <typename T, bool = std::is_move_constructible<T>::value&&
+                          std::is_move_assignable<T>::value>
+struct MoveAssignBase {
+  MoveAssignBase() = default;
+  MoveAssignBase(const MoveAssignBase&) = default;
+  MoveAssignBase(MoveAssignBase&&) = default;
+  MoveAssignBase& operator=(const MoveAssignBase&) = default;
+  MoveAssignBase& operator=(MoveAssignBase&&) = default;
+};
+
+template <typename T>
+struct MoveAssignBase<T, false> {
+  MoveAssignBase() = default;
+  MoveAssignBase(const MoveAssignBase&) = default;
+  MoveAssignBase(MoveAssignBase&&) = default;
+  MoveAssignBase& operator=(const MoveAssignBase&) = default;
+  MoveAssignBase& operator=(MoveAssignBase&&) = delete;
+};
+
+void ThrowBadStatusOrAccess(absl::Status status);
+
+}  // namespace internal_statusor
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_STATUSOR_INTERNALS_H_
diff --git a/tensorflow_lite_support/cc/port/default/tflite_wrapper.cc b/tensorflow_lite_support/cc/port/default/tflite_wrapper.cc
new file mode 100644
index 0000000..548e679
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/default/tflite_wrapper.cc
@@ -0,0 +1,60 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/port/default/tflite_wrapper.h"
+
+#include "absl/status/status.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+
+namespace tflite {
+namespace support {
+
+absl::Status TfLiteInterpreterWrapper::InitializeWithFallback(
+    std::function<absl::Status(std::unique_ptr<tflite::Interpreter>*)>
+        interpreter_initializer,
+    const tflite::proto::ComputeSettings& compute_settings) {
+  if (compute_settings.has_preference() ||
+      compute_settings.has_tflite_settings()) {
+    return absl::UnimplementedError(
+        "Acceleration via ComputeSettings is not supported yet.");
+  }
+  RETURN_IF_ERROR(interpreter_initializer(&interpreter_));
+  return interpreter_->AllocateTensors() != kTfLiteOk
+             ? absl::InternalError(
+                   "TFLite interpreter: AllocateTensors() failed.")
+             : absl::OkStatus();
+}
+
+absl::Status TfLiteInterpreterWrapper::InvokeWithFallback(
+    const std::function<absl::Status(tflite::Interpreter* interpreter)>&
+        set_inputs) {
+  RETURN_IF_ERROR(set_inputs(interpreter_.get()));
+  return interpreter_->Invoke() != kTfLiteOk
+             ? absl::InternalError("TFLite interpreter: Invoke() failed.")
+             : absl::OkStatus();
+}
+
+absl::Status TfLiteInterpreterWrapper::InvokeWithoutFallback() {
+  return interpreter_->Invoke() != kTfLiteOk
+             ? absl::InternalError("TFLite interpreter: Invoke() failed.")
+             : absl::OkStatus();
+}
+
+void TfLiteInterpreterWrapper::Cancel() {
+  // NOP
+}
+
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/port/default/tflite_wrapper.h b/tensorflow_lite_support/cc/port/default/tflite_wrapper.h
new file mode 100644
index 0000000..3fd489f
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/default/tflite_wrapper.h
@@ -0,0 +1,82 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_TFLITE_WRAPPER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_TFLITE_WRAPPER_H_
+
+#include <memory>
+#include <utility>
+
+#include "absl/status/status.h"
+#include "tensorflow/lite/experimental/acceleration/configuration/configuration.pb.h"
+#include "tensorflow/lite/interpreter.h"
+
+namespace tflite {
+namespace support {
+
+// Wrapper for a TfLiteInterpreter that may be accelerated[1]. This is NOT yet
+// implemented: this class only provides a first, minimal interface in the
+// meanwhile.
+//
+// [1] See tensorflow/lite/experimental/acceleration for more details.
+class TfLiteInterpreterWrapper {
+ public:
+  TfLiteInterpreterWrapper() = default;
+
+  virtual ~TfLiteInterpreterWrapper() = default;
+
+  // Calls `interpreter_initializer` and then `AllocateTensors`. Future
+  // implementation of this method will attempt to apply the provided
+  // `compute_settings` with a graceful fallback in case a failure occurs.
+  // Note: before this gets implemented, do NOT call this method with non-empty
+  // `compute_settings` otherwise an unimplemented error occurs.
+  absl::Status InitializeWithFallback(
+      std::function<absl::Status(std::unique_ptr<tflite::Interpreter>*)>
+          interpreter_initializer,
+      const tflite::proto::ComputeSettings& compute_settings);
+
+  // Calls `set_inputs` and then Invoke() on the interpreter. Future
+  // implementation of this method will perform a graceful fallback in case a
+  // failure occur due to the `compute_settings` provided at initialization
+  // time.
+  absl::Status InvokeWithFallback(
+      const std::function<absl::Status(tflite::Interpreter* interpreter)>&
+          set_inputs);
+
+  // Calls Invoke() on the interpreter. Caller must have set up inputs
+  // before-hand.
+  absl::Status InvokeWithoutFallback();
+
+  // Cancels the current running TFLite invocation on CPU. This method is not
+  // yet implemented though it is safe to use it as it acts as a NOP.
+  void Cancel();
+
+  // Accesses the underlying interpreter for other methods.
+  tflite::Interpreter& operator*() { return *interpreter_; }
+  tflite::Interpreter* operator->() { return interpreter_.get(); }
+  tflite::Interpreter& operator*() const { return *interpreter_; }
+  tflite::Interpreter* operator->() const { return interpreter_.get(); }
+  tflite::Interpreter* get() const { return interpreter_.get(); }
+
+  TfLiteInterpreterWrapper(const TfLiteInterpreterWrapper&) = delete;
+  TfLiteInterpreterWrapper& operator=(const TfLiteInterpreterWrapper&) = delete;
+
+ private:
+  std::unique_ptr<tflite::Interpreter> interpreter_;
+};
+
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_DEFAULT_TFLITE_WRAPPER_H_
diff --git a/tensorflow_lite_support/cc/port/gmock.h b/tensorflow_lite_support/cc/port/gmock.h
new file mode 100644
index 0000000..5e4334d
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/gmock.h
@@ -0,0 +1,21 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_GMOCK_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_GMOCK_H_
+
+#include "gmock/gmock.h"
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_GMOCK_H_
diff --git a/tensorflow_lite_support/cc/port/gtest.h b/tensorflow_lite_support/cc/port/gtest.h
new file mode 100644
index 0000000..dbe2e5e
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/gtest.h
@@ -0,0 +1,21 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_GTEST_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_GTEST_H_
+
+#include "gtest/gtest.h"
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_GTEST_H_
diff --git a/tensorflow_lite_support/cc/port/integral_types.h b/tensorflow_lite_support/cc/port/integral_types.h
new file mode 100644
index 0000000..76d9d50
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/integral_types.h
@@ -0,0 +1,46 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_INTEGRAL_TYPES_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_INTEGRAL_TYPES_H_
+
+// Add namespace here to avoid conflict with other libraries.
+namespace tflite {
+
+typedef signed char schar;
+typedef signed char int8;
+typedef short int16;
+typedef int int32;
+typedef long long int64;
+
+typedef unsigned char uint8;
+typedef unsigned short uint16;
+typedef unsigned int uint32;
+typedef unsigned int char32;
+typedef unsigned long long uint64;
+typedef unsigned long uword_t;
+
+#define GG_LONGLONG(x) x##LL
+#define GG_ULONGLONG(x) x##ULL
+#define GG_LL_FORMAT "ll"  // As in "%lld". Note that "q" is poor form also.
+#define GG_LL_FORMAT_W L"ll"
+
+typedef uint64 Fprint;
+static const Fprint kIllegalFprint = 0;
+static const Fprint kMaxFprint = GG_ULONGLONG(0xFFFFFFFFFFFFFFFF);
+
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_INTEGRAL_TYPES_H_
diff --git a/tensorflow_lite_support/cc/port/status_macros.h b/tensorflow_lite_support/cc/port/status_macros.h
new file mode 100644
index 0000000..3890c77
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/status_macros.h
@@ -0,0 +1,21 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_STATUS_MACROS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_STATUS_MACROS_H_
+
+#include "tensorflow_lite_support/cc/port/default/status_macros.h"
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_STATUS_MACROS_H_
diff --git a/tensorflow_lite_support/cc/port/statusor.h b/tensorflow_lite_support/cc/port/statusor.h
new file mode 100644
index 0000000..f84c756
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/statusor.h
@@ -0,0 +1,20 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_STATUSOR_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_STATUSOR_H_
+
+#include "tensorflow_lite_support/cc/port/default/statusor.h"
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_STATUSOR_H_
diff --git a/tensorflow_lite_support/cc/port/tflite_wrapper.h b/tensorflow_lite_support/cc/port/tflite_wrapper.h
new file mode 100644
index 0000000..601df9b
--- /dev/null
+++ b/tensorflow_lite_support/cc/port/tflite_wrapper.h
@@ -0,0 +1,21 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_PORT_TFLITE_WRAPPER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_PORT_TFLITE_WRAPPER_H_
+
+#include "tensorflow_lite_support/cc/port/default/tflite_wrapper.h"
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_PORT_TFLITE_WRAPPER_H_
diff --git a/tensorflow_lite_support/cc/task/README.md b/tensorflow_lite_support/cc/task/README.md
new file mode 100644
index 0000000..bd756a2
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/README.md
@@ -0,0 +1,384 @@
+# TFLite Task library - C++
+
+A flexible and ready-to-use library for common machine learning model types,
+such as classification and detection.
+
+## Text Task Libraries
+
+### QuestionAnswerer
+
+`QuestionAnswerer` API is able to load
+[Mobile BERT](https://tfhub.dev/tensorflow/mobilebert/1) or
+[AlBert](https://tfhub.dev/tensorflow/albert_lite_base/1) TFLite models and
+answer question based on context.
+
+Use the C++ API to answer questions as follows:
+
+```cc
+using tflite::task::text::qa::BertQuestionAnswerer;
+using tflite::task::text::qa::QaAnswer;
+// Create API handler with Mobile Bert model.
+auto qa_client = BertQuestionAnswerer::CreateBertQuestionAnswererFromFile("/path/to/mobileBertModel", "/path/to/vocab");
+// Or create API handler with Albert model.
+// auto qa_client = BertQuestionAnswerer::CreateAlbertQuestionAnswererFromFile("/path/to/alBertModel", "/path/to/sentencePieceModel");
+
+
+std::string context =
+    "Nikola Tesla (Serbian Cyrillic: Никола Тесла; 10 "
+    "July 1856 – 7 January 1943) was a Serbian American inventor, electrical "
+    "engineer, mechanical engineer, physicist, and futurist best known for his "
+    "contributions to the design of the modern alternating current (AC) "
+    "electricity supply system.";
+std::string question = "When was Nikola Tesla born?";
+// Run inference with `context` and a given `question` to the context, and get top-k
+// answers ranked by logits.
+const std::vector<QaAnswer> answers = qa_client->Answer(context, question);
+// Access QaAnswer results.
+for (const QaAnswer& item : answers) {
+  std::cout << absl::StrFormat("Text: %s logit=%f start=%d end=%d", item.text,
+                               item.pos.logit, item.pos.start, item.pos.end)
+            << std::endl;
+}
+// Output:
+// Text: 10 July 1856 logit=16.8527 start=17 end=19
+// ... (and more)
+//
+// So the top-1 answer is: "10 July 1856".
+```
+
+In the above code, `item.text` is the text content of an answer. We use a span
+with closed interval `[item.pos.start, item.pos.end]` to denote predicted tokens
+in the answer, and `item.pos.logit` is the sum of span logits to represent the
+confidence score.
+
+### NLClassifier
+
+`NLClassifier` API is able to load any TFLite models for natural language
+classaification task such as language detection or sentiment detection.
+
+The API expects a TFLite model with the following input/output tensor:
+Input tensor0:
+  (kTfLiteString) - input of the model, accepts a string.
+Output tensor0:
+  (kTfLiteUInt8/kTfLiteInt8/kTfLiteInt16/kTfLiteFloat32/kTfLiteFloat64)
+  - output scores for each class, if type is one of the Int types,
+    dequantize it to double
+Output tensor1: optional
+  (kTfLiteString)
+  - output classname for each class, should be of the same length with
+    scores. If this tensor is not present, the API uses score indices as
+    classnames.
+By default the API tries to find the input/output tensors with default
+configurations in NLClassifierOptions, with tensor name prioritized over
+tensor index. The option is configurable for different TFLite models.
+
+Use the C++ API to perform language ID classification as follows:
+
+```cc
+using tflite::task::text::nlclassifier::NLClassifier;
+using tflite::task::core::Category;
+auto classifier = NLClassifier::CreateFromFileAndOptions("/path/to/model");
+// Or create a customized NLClassifierOptions
+// NLClassifierOptions options =
+//   {
+//     .output_score_tensor_name = myOutputScoreTensorName,
+//     .output_label_tensor_name = myOutputLabelTensorName,
+//   }
+// auto classifier = NLClassifier::CreateFromFileAndOptions("/path/to/model", options);
+std::string context = "What language is this?";
+std::vector<Category> categories = classifier->Classify(context);
+// Access category results.
+for (const Categoryr& category : categories) {
+  std::cout << absl::StrFormat("Language: %s Probability: %f", category.class_name, category_.score)
+            << std::endl;
+}
+// Output:
+// Language: en Probability=0.9
+// ... (and more)
+//
+// So the top-1 answer is 'en'.
+```
+
+## Vision Task Libraries
+
+### Image Classifier
+
+`ImageClassifier` accepts any TFLite image classification model (with optional,
+but strongly recommended, TFLite Model Metadata) that conforms to the following
+spec:
+
+Input tensor (type: `kTfLiteUInt8` / `kTfLiteFloat32`):
+
+   - image input of size `[batch x height x width x channels]`.
+   - batch inference is not supported (`batch` is required to be 1).
+   - only RGB inputs are supported (`channels` is required to be 3).
+   - if type is `kTfLiteFloat32`, `NormalizationOptions` are required to be
+     attached to the metadata for input normalization.
+
+At least one output tensor (type: `kTfLiteUInt8` / `kTfLiteFloat32`) with:
+
+   -  `N` classes and either 2 or 4 dimensions, i.e. `[1 x N]` or
+      `[1 x 1 x 1 x N]`
+   - optional (but recommended) label map(s) as AssociatedFile-s with type
+     TENSOR_AXIS_LABELS, containing one label per line. The first such
+     AssociatedFile (if any) is used to fill the `class_name` field of the
+     results. The `display_name` field is filled from the AssociatedFile (if
+     any) whose locale matches the `display_names_locale` field of the
+     `ImageClassifierOptions` used at creation time ("en" by default, i.e.
+     English). If none of these are available, only the `index` field of the
+     results will be filled.
+
+An example of such model can be found at:
+https://tfhub.dev/bohemian-visual-recognition-alliance/lite-model/models/mushroom-identification_v1/1
+
+Example usage:
+
+```cc
+// More options are available (e.g. max number of results to return). At the
+// very least, the model must be specified:
+ImageClassifierOptions options;
+options.mutable_model_file_with_metadata()->set_file_name(
+    "/path/to/model.tflite");
+
+// Create an ImageClassifier instance from the options.
+StatusOr<std::unique_ptr<ImageClassifier>> image_classifier_or =
+    ImageClassifier::CreateFromOptions(options);
+// Check if an error occurred.
+if (!image_classifier_or.ok()) {
+  std::cerr << "An error occurred during ImageClassifier creation: "
+            << image_classifier_or.status().message();
+  return;
+}
+std::unique_ptr<ImageClassifier> image_classifier =
+    std::move(image_classifier_or.value());
+
+// Prepare FrameBuffer input from e.g. image RGBA data, width and height:
+std::unique_ptr<FrameBuffer> frame_buffer =
+    CreateFromRgbaRawBuffer(image_rgba_data, {image_width, image_height});
+
+// Run inference:
+StatusOr<ClassificationResult> result_or =
+    image_classifier->Classify(*frame_buffer);
+// Check if an error occurred.
+if (!result_or.ok()) {
+  std::cerr << "An error occurred during classification: "
+            << result_or.status().message();
+  return;
+}
+ClassificationResult result = result_or.value();
+
+// Example value for 'result':
+//
+// classifications {
+//   classes { index: 934 score: 0.95 class_name: "cat" }
+//   classes { index: 948 score: 0.007 class_name: "dog" }
+//   classes { index: 927 score: 0.003 class_name: "fox" }
+//   head_index: 0
+// }
+```
+
+A CLI demo tool is also available [here][1] for easily trying out this API.
+
+### Object Detector
+
+`ObjectDetector` accepts any object detection TFLite model (with mandatory
+TFLite Model Metadata) that conforms to the following spec (e.g. Single Shot
+Detectors):
+
+Input tensor (type: `kTfLiteUInt8` / `kTfLiteFloat32`):
+
+   - image input of size `[batch x height x width x channels]`.
+   - batch inference is not supported (`batch` is required to be 1).
+   - only RGB inputs are supported (`channels` is required to be 3).
+   - if type is kTfLiteFloat32, `NormalizationOptions` are required to be
+     attached to the metadata for input normalization.
+
+Output tensors must be the 4 outputs (type: `kTfLiteFloat32`) of a
+[`DetectionPostProcess`][2] op, i.e:
+
+* Locations:
+
+  - of size `[num_results x 4]`, the inner array
+    representing bounding boxes in the form [top, left, right, bottom].
+  - BoundingBoxProperties are required to be attached to the metadata
+    and must specify type=BOUNDARIES and coordinate_type=RATIO.
+
+* Classes:
+
+  - of size `[num_results]`, each value representing the
+    integer index of a class.
+  - optional (but recommended) label map(s) can be attached as
+    AssociatedFile-s with type TENSOR_VALUE_LABELS, containing one label per
+    line. The first such AssociatedFile (if any) is used to fill the
+    `class_name` field of the results. The `display_name` field is filled
+    from the AssociatedFile (if any) whose locale matches the
+    `display_names_locale` field of the `ObjectDetectorOptions` used at
+    creation time ("en" by default, i.e. English). If none of these are
+    available, only the `index` field of the results will be filled.
+
+* Scores:
+
+  - of size `[num_results]`, each value representing the score
+    of the detected object.
+
+* Number of results:
+
+  - integer `num_results` as a tensor of size `[1]`
+
+An example of such model can be found at:
+https://tfhub.dev/google/lite-model/object_detection/mobile_object_localizer_v1/1/metadata/1
+
+Example usage:
+
+```cc
+// More options are available (e.g. max number of results to return). At the
+// very least, the model must be specified:
+ObjectDetectorOptions options;
+options.mutable_model_file_with_metadata()->set_file_name(
+    "/path/to/model.tflite");
+
+// Create an ObjectDetector instance from the options.
+StatusOr<std::unique_ptr<ObjectDetector>> object_detector_or =
+    ObjectDetector::CreateFromOptions(options);
+// Check if an error occurred.
+if (!object_detector_or.ok()) {
+  std::cerr << "An error occurred during ObjectDetector creation: "
+            << object_detector_or.status().message();
+  return;
+}
+std::unique_ptr<ObjectDetector> object_detector =
+    std::move(object_detector_or.value());
+
+// Prepare FrameBuffer input from e.g. image RGBA data, width and height:
+std::unique_ptr<FrameBuffer> frame_buffer =
+    CreateFromRgbaRawBuffer(image_rgba_data, {image_width, image_height});
+
+// Run inference:
+StatusOr<DetectionResult> result_or = object_detector->Detect(*frame_buffer);
+// Check if an error occurred.
+if (!result_or.ok()) {
+  std::cerr << "An error occurred during detection: "
+            << result_or.status().message();
+  return;
+}
+DetectionResult result = result_or.value();
+
+// Example value for 'result':
+//
+// detections {
+//   bounding_box {
+//     origin_x: 54
+//     origin_y: 398
+//     width: 393
+//     height: 196
+//   }
+//   classes { index: 16 score: 0.65 class_name: "cat" }
+// }
+// detections {
+//   bounding_box {
+//     origin_x: 602
+//     origin_y: 157
+//     width: 394
+//     height: 447
+//   }
+//   classes { index: 17 score: 0.45 class_name: "dog" }
+// }
+```
+
+A CLI demo tool is available [here][3] for easily trying out this API.
+
+### Image Segmenter
+
+`ImageSegmenter` accepts any TFLite model (with optional, but strongly
+recommended, TFLite Model Metadata) that conforms to the following spec:
+
+Input tensor (type: `kTfLiteUInt8` / `kTfLiteFloat32`):
+
+   - image input of size `[batch x height x width x channels]`.
+   - batch inference is not supported (`batch` is required to be 1).
+   - only RGB inputs are supported (`channels` is required to be 3).
+   - if type is kTfLiteFloat32, `NormalizationOptions` are required to be
+     attached to the metadata for input normalization.
+
+Output tensor (type: `kTfLiteUInt8` / `kTfLiteFloat32`):
+
+   - tensor of size `[batch x mask_height x mask_width x num_classes]`, where
+     `batch` is required to be 1, `mask_width` and `mask_height` are the
+     dimensions of the segmentation masks produced by the model, and
+     `num_classes` is the number of classes supported by the model.
+   - optional (but recommended) label map(s) can be attached as
+     AssociatedFile-s with type TENSOR_AXIS_LABELS, containing one label per
+     line. The first such AssociatedFile (if any) is used to fill the
+     `class_name` field of the results. The `display_name` field is filled
+     from the AssociatedFile (if any) whose locale matches the
+     `display_names_locale` field of the `ImageSegmenterOptions` used at
+     creation time ("en" by default, i.e. English). If none of these are
+     available, only the `index` field of the results will be filled.
+
+An example of such model can be found at:
+https://tfhub.dev/tensorflow/lite-model/deeplabv3/1/metadata/1
+
+Example usage:
+
+```cc
+// More options are available to select between return a single category mask
+// or multiple confidence masks during post-processing.
+ImageSegmenterOptions options;
+options.mutable_model_file_with_metadata()->set_file_name(
+    "/path/to/model.tflite");
+
+// Create an ImageSegmenter instance from the options.
+StatusOr<std::unique_ptr<ImageSegmenter>> image_segmenter_or =
+    ImageSegmenter::CreateFromOptions(options);
+// Check if an error occurred.
+if (!image_segmenter_or.ok()) {
+  std::cerr << "An error occurred during ImageSegmenter creation: "
+            << image_segmenter_or.status().message();
+  return;
+}
+std::unique_ptr<ImageSegmenter> immage_segmenter =
+    std::move(image_segmenter_or.value());
+
+// Prepare FrameBuffer input from e.g. image RGBA data, width and height:
+std::unique_ptr<FrameBuffer> frame_buffer =
+    CreateFromRgbaRawBuffer(image_rgba_data, {image_width, image_height});
+
+// Run inference:
+StatusOr<SegmentationResult> result_or =
+    immage_segmenter->Segment(*frame_buffer);
+// Check if an error occurred.
+if (!result_or.ok()) {
+  std::cerr << "An error occurred during segmentation: "
+            << result_or.status().message();
+  return;
+}
+SegmentationResult result = result_or.value();
+
+// Example value for 'result':
+//
+// segmentation {
+//   width: 257
+//   height: 257
+//   category_mask: "\x00\x01..."
+//   colored_labels { r: 0 g: 0 b: 0 class_name: "background" }
+//   colored_labels { r: 128 g: 0 b: 0 class_name: "aeroplane" }
+//   ...
+//   colored_labels { r: 128 g: 192 b: 0 class_name: "train" }
+//   colored_labels { r: 0 g: 64 b: 128 class_name: "tv" }
+// }
+//
+// Where 'category_mask' is a byte buffer of size 'width' x 'height', with the
+// value of each pixel representing the class this pixel belongs to (e.g. '\x00'
+// means "background", '\x01' means "aeroplane", etc).
+// 'colored_labels' provides the label for each possible value, as well as
+// suggested RGB components to optionally transform the result into a more
+// human-friendly colored image.
+//
+```
+
+A CLI demo tool is available [here][4] for easily trying out this API.
+
+[1]: https://github.com/tensorflow/tflite-support/blob/master/tensorflow_lite_support/examples/task/vision/desktop/image_classifier_demo.cc
+[2]: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/kernels/detection_postprocess.cc
+[3]: https://github.com/tensorflow/tflite-support/blob/master/tensorflow_lite_support/examples/task/vision/desktop/object_detector_demo.cc
+[4]: https://github.com/tensorflow/tflite-support/blob/master/tensorflow_lite_support/examples/task/vision/desktop/image_segmenter_demo.cc
diff --git a/tensorflow_lite_support/cc/task/core/BUILD b/tensorflow_lite_support/cc/task/core/BUILD
new file mode 100644
index 0000000..1995dfe
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/BUILD
@@ -0,0 +1,156 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "tflite_engine",
+    srcs = ["tflite_engine.cc"],
+    hdrs = ["tflite_engine.h"],
+    deps = [
+        ":external_file_handler",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api:op_resolver",
+        # The dependency on builtin_ops here is only for the default
+        # value of the OpResolver parameter:
+        #     std::unique_ptr<tflite::IterableOpResolver> resolver =
+        #       absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>()
+        # When linking statically, if the client of this library doesn't use
+        # the default argument, this dependency does not cause all the builtin ops
+        # to get included in the executable.
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+        "@org_tensorflow//tensorflow/lite/tools:verifier",
+    ] + select({
+        "//tensorflow_lite_support/cc:tflite_use_c_api": [
+            "@org_tensorflow//tensorflow/lite/core/api:verifier",
+            "@org_tensorflow//tensorflow/lite/c:c_api",
+            "@org_tensorflow//tensorflow/lite/c:c_api_experimental",
+            "@org_tensorflow//tensorflow/lite:kernel_api",
+            "@org_tensorflow//tensorflow/lite:stderr_reporter",
+        ],
+        "//conditions:default": [
+            "@org_tensorflow//tensorflow/lite:framework",
+            "@org_tensorflow//tensorflow/lite:kernel_api",
+        ],
+    }) + [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:tflite_wrapper",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+    ],
+)
+
+# This is a duplicate of the above 'tflite_engine' target that is used for
+# testing with TFLITE_USE_C_API defined.  It should be the same as the target
+# above, except that it adds
+#    testonly = 1,
+#    defines = ["TFLITE_USE_C_API"],
+# and that it resolves the conditional deps from the 'select' as if
+# "//tensorflow_lite_support/cc:tflite_use_c_api" was enabled.
+# This allows testing the TFLITE_USE_C_API case even when
+# '--copt=-DTFLITE_USE_C_API' wasn't passed on the build command line.
+cc_library(
+    name = "tflite_engine_with_c_api_for_test",
+    testonly = 1,
+    srcs = ["tflite_engine.cc"],
+    hdrs = ["tflite_engine.h"],
+    defines = ["TFLITE_USE_C_API"],
+    deps = [
+        ":external_file_handler",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+        "@org_tensorflow//tensorflow/lite/tools:verifier",
+    ] + [
+        "@org_tensorflow//tensorflow/lite/core/api:op_resolver",
+        "@org_tensorflow//tensorflow/lite/core/api:verifier",
+        "@org_tensorflow//tensorflow/lite/c:c_api",
+        "@org_tensorflow//tensorflow/lite/c:c_api_experimental",
+        "@org_tensorflow//tensorflow/lite:kernel_api",
+        "@org_tensorflow//tensorflow/lite:stderr_reporter",
+    ] + [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:tflite_wrapper_with_c_api_for_test",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+    ],
+)
+
+cc_library(
+    name = "base_task_api",
+    hdrs = ["base_task_api.h"],
+    deps = [
+        ":tflite_engine",
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/port:tflite_wrapper",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@org_tensorflow//tensorflow/lite/c:common",
+    ],
+)
+
+cc_library(
+    name = "task_api_factory",
+    hdrs = ["task_api_factory.h"],
+    deps = [
+        ":base_task_api",
+        ":tflite_engine",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+        "@com_google_absl//absl/status",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api:op_resolver",
+        "@org_tensorflow//tensorflow/lite/kernels:op_macros",
+    ],
+)
+
+cc_library(
+    name = "task_utils",
+    srcs = ["task_utils.cc"],
+    hdrs = ["task_utils.h"],
+    deps = [
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/strings",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:string_util",
+        "@org_tensorflow//tensorflow/lite:type_to_tflitetype",
+        "@org_tensorflow//tensorflow/lite/kernels:op_macros",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+    ],
+)
+
+cc_library(
+    name = "category",
+    hdrs = ["category.h"],
+)
+
+cc_library(
+    name = "external_file_handler",
+    srcs = ["external_file_handler.cc"],
+    hdrs = ["external_file_handler.h"],
+    visibility = ["//visibility:public"],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+        "@com_google_absl//absl/base:core_headers",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/task/core/base_task_api.h b/tensorflow_lite_support/cc/task/core/base_task_api.h
new file mode 100644
index 0000000..a27f785
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/base_task_api.h
@@ -0,0 +1,144 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_BASE_TASK_API_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_BASE_TASK_API_H_
+
+#include <utility>
+
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/port/tflite_wrapper.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+
+namespace tflite {
+namespace task {
+namespace core {
+
+class BaseUntypedTaskApi {
+ public:
+  explicit BaseUntypedTaskApi(std::unique_ptr<TfLiteEngine> engine)
+      : engine_{std::move(engine)} {}
+
+  virtual ~BaseUntypedTaskApi() = default;
+
+  TfLiteEngine* GetTfLiteEngine() { return engine_.get(); }
+  const TfLiteEngine* GetTfLiteEngine() const { return engine_.get(); }
+
+  const metadata::ModelMetadataExtractor* GetMetadataExtractor() const {
+    return engine_->metadata_extractor();
+  }
+
+ protected:
+  std::unique_ptr<TfLiteEngine> engine_;
+};
+
+template <class OutputType, class... InputTypes>
+class BaseTaskApi : public BaseUntypedTaskApi {
+ public:
+  explicit BaseTaskApi(std::unique_ptr<TfLiteEngine> engine)
+      : BaseUntypedTaskApi(std::move(engine)) {}
+  // BaseTaskApi is neither copyable nor movable.
+  BaseTaskApi(const BaseTaskApi&) = delete;
+  BaseTaskApi& operator=(const BaseTaskApi&) = delete;
+
+  // Cancels the current running TFLite invocation on CPU.
+  //
+  // Usually called on a different thread than the one inference is running on.
+  // Calling Cancel() will cause the underlying TFLite interpreter to return an
+  // error, which will turn into a `CANCELLED` status and empty results. Calling
+  // Cancel() at the other time will not take any effect on the current or
+  // following invocation. It is perfectly fine to run inference again on the
+  // same instance after a cancelled invocation. If the TFLite inference is
+  // partially delegated on CPU, logs a warning message and only cancels the
+  // invocation running on CPU. Other invocation which depends on the output of
+  // the CPU invocation will not be executed.
+  void Cancel() { engine_->Cancel(); }
+
+ protected:
+  // Subclasses need to populate input_tensors from api_inputs.
+  virtual absl::Status Preprocess(
+      const std::vector<TfLiteTensor*>& input_tensors,
+      InputTypes... api_inputs) = 0;
+
+  // Subclasses need to construct OutputType object from output_tensors.
+  // Original inputs are also provided as they may be needed.
+  virtual tflite::support::StatusOr<OutputType> Postprocess(
+      const std::vector<const TfLiteTensor*>& output_tensors,
+      InputTypes... api_inputs) = 0;
+
+  // Returns (the addresses of) the model's inputs.
+  std::vector<TfLiteTensor*> GetInputTensors() { return engine_->GetInputs(); }
+
+  // Returns (the addresses of) the model's outputs.
+  std::vector<const TfLiteTensor*> GetOutputTensors() {
+    return engine_->GetOutputs();
+  }
+
+  // Performs inference using tflite::support::TfLiteInterpreterWrapper
+  // InvokeWithoutFallback().
+  tflite::support::StatusOr<OutputType> Infer(InputTypes... args) {
+    tflite::task::core::TfLiteEngine::InterpreterWrapper* interpreter_wrapper =
+        engine_->interpreter_wrapper();
+    // Note: AllocateTensors() is already performed by the interpreter wrapper
+    // at InitInterpreter time (see TfLiteEngine).
+    RETURN_IF_ERROR(Preprocess(GetInputTensors(), args...));
+    absl::Status status = interpreter_wrapper->InvokeWithoutFallback();
+    if (!status.ok()) {
+      return status.GetPayload(tflite::support::kTfLiteSupportPayload)
+                     .has_value()
+                 ? status
+                 : tflite::support::CreateStatusWithPayload(status.code(),
+                                                            status.message());
+    }
+    return Postprocess(GetOutputTensors(), args...);
+  }
+
+  // Performs inference using tflite::support::TfLiteInterpreterWrapper
+  // InvokeWithFallback() to benefit from automatic fallback from delegation to
+  // CPU where applicable.
+  tflite::support::StatusOr<OutputType> InferWithFallback(InputTypes... args) {
+    tflite::task::core::TfLiteEngine::InterpreterWrapper* interpreter_wrapper =
+        engine_->interpreter_wrapper();
+    // Note: AllocateTensors() is already performed by the interpreter wrapper
+    // at InitInterpreter time (see TfLiteEngine).
+    RETURN_IF_ERROR(Preprocess(GetInputTensors(), args...));
+    auto set_inputs_nop =
+        [](tflite::task::core::TfLiteEngine::Interpreter* interpreter)
+        -> absl::Status {
+      // NOP since inputs are populated at Preprocess() time.
+      return absl::OkStatus();
+    };
+    absl::Status status =
+        interpreter_wrapper->InvokeWithFallback(set_inputs_nop);
+    if (!status.ok()) {
+      return status.GetPayload(tflite::support::kTfLiteSupportPayload)
+                     .has_value()
+                 ? status
+                 : tflite::support::CreateStatusWithPayload(status.code(),
+                                                            status.message());
+    }
+    return Postprocess(GetOutputTensors(), args...);
+  }
+};
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_BASE_TASK_API_H_
diff --git a/tensorflow_lite_support/cc/task/core/category.h b/tensorflow_lite_support/cc/task/core/category.h
new file mode 100644
index 0000000..a99f994
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/category.h
@@ -0,0 +1,44 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_CATEGORY_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_CATEGORY_H_
+#include <string>
+
+namespace tflite {
+namespace task {
+namespace core {
+
+// Result for classification APIs.
+struct Category {
+  std::string class_name;
+  double score;
+  Category(const std::string& class_name, double score)
+      : class_name(class_name), score(score) {}
+
+  friend bool operator==(const Category& lhs, const Category& rhs) {
+    return lhs.score == rhs.score && lhs.class_name == rhs.class_name;
+  }
+
+  friend bool operator!=(const Category& lhs, const Category& rhs) {
+    return !(lhs == rhs);
+  }
+};
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_CATEGORY_H_
diff --git a/tensorflow_lite_support/cc/task/core/external_file_handler.cc b/tensorflow_lite_support/cc/task/core/external_file_handler.cc
new file mode 100644
index 0000000..e2150c1
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/external_file_handler.cc
@@ -0,0 +1,194 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stddef.h>
+#include <sys/mman.h>
+#include <unistd.h>
+
+#include <memory>
+#include <string>
+
+#include "absl/memory/memory.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+
+namespace tflite {
+namespace task {
+namespace core {
+namespace {
+
+using ::absl::StatusCode;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+
+// Gets the offset aligned to page size for mapping given files into memory by
+// file descriptor correctly, as according to mmap(2), the offset used in mmap
+// must be a multiple of sysconf(_SC_PAGE_SIZE).
+int64 GetPageSizeAlignedOffset(int64 offset) {
+  int64 aligned_offset = offset;
+  int64 page_size = sysconf(_SC_PAGE_SIZE);
+  if (offset % page_size != 0) {
+    aligned_offset = offset / page_size * page_size;
+  }
+  return aligned_offset;
+}
+
+}  // namespace
+
+/* static */
+StatusOr<std::unique_ptr<ExternalFileHandler>>
+ExternalFileHandler::CreateFromExternalFile(const ExternalFile* external_file) {
+  // Use absl::WrapUnique() to call private constructor:
+  // https://abseil.io/tips/126.
+  std::unique_ptr<ExternalFileHandler> handler =
+      absl::WrapUnique(new ExternalFileHandler(external_file));
+
+  RETURN_IF_ERROR(handler->MapExternalFile());
+
+  return handler;
+}
+
+absl::Status ExternalFileHandler::MapExternalFile() {
+  if (!external_file_.file_content().empty()) {
+    return absl::OkStatus();
+  }
+  if (external_file_.file_name().empty() &&
+      !external_file_.has_file_descriptor_meta()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "ExternalFile must specify at least one of 'file_content', file_name' "
+        "or 'file_descriptor_meta'.",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  // Obtain file descriptor, offset and size.
+  int fd = -1;
+  if (!external_file_.file_name().empty()) {
+    owned_fd_ = open(external_file_.file_name().c_str(), O_RDONLY);
+    if (owned_fd_ < 0) {
+      const std::string error_message = absl::StrFormat(
+          "Unable to open file at %s", external_file_.file_name());
+      switch (errno) {
+        case ENOENT:
+          return CreateStatusWithPayload(
+              StatusCode::kNotFound, error_message,
+              TfLiteSupportStatus::kFileNotFoundError);
+        case EACCES:
+        case EPERM:
+          return CreateStatusWithPayload(
+              StatusCode::kPermissionDenied, error_message,
+              TfLiteSupportStatus::kFilePermissionDeniedError);
+        case EINTR:
+          return CreateStatusWithPayload(StatusCode::kUnavailable,
+                                         error_message,
+                                         TfLiteSupportStatus::kFileReadError);
+        case EBADF:
+          return CreateStatusWithPayload(StatusCode::kFailedPrecondition,
+                                         error_message,
+                                         TfLiteSupportStatus::kFileReadError);
+        default:
+          return CreateStatusWithPayload(
+              StatusCode::kUnknown,
+              absl::StrFormat("%s, errno=%d", error_message, errno),
+              TfLiteSupportStatus::kFileReadError);
+      }
+    }
+    fd = owned_fd_;
+  } else {
+    fd = external_file_.file_descriptor_meta().fd();
+    if (fd < 0) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Provided file descriptor is invalid: %d < 0", fd),
+          TfLiteSupportStatus::kInvalidArgumentError);
+    }
+    buffer_offset_ = external_file_.file_descriptor_meta().offset();
+    buffer_size_ = external_file_.file_descriptor_meta().length();
+  }
+  // Get actual file size. Always use 0 as offset to lseek(2) to get the actual
+  // file size, as SEEK_END returns the size of the file *plus* offset.
+  size_t file_size = lseek(fd, /*offset=*/0, SEEK_END);
+  if (file_size <= 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown,
+        absl::StrFormat("Unable to get file size, errno=%d", errno),
+        TfLiteSupportStatus::kFileReadError);
+  }
+  // Deduce buffer size if not explicitly provided through file descriptor.
+  if (buffer_size_ <= 0) {
+    buffer_size_ = file_size - buffer_offset_;
+  }
+  // Check for out of range issues.
+  if (file_size <= buffer_offset_) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Provided file offset (%d) exceeds or matches actual "
+                        "file length (%d)",
+                        buffer_offset_, file_size),
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (file_size < buffer_size_ + buffer_offset_) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Provided file length + offset (%d) exceeds actual "
+                        "file length (%d)",
+                        buffer_size_ + buffer_offset_, file_size),
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  // If buffer_offset_ is not multiple of sysconf(_SC_PAGE_SIZE), align with
+  // extra leading bytes and adjust buffer_size_ to account for the extra
+  // leading bytes.
+  buffer_aligned_offset_ = GetPageSizeAlignedOffset(buffer_offset_);
+  buffer_aligned_size_ = buffer_size_ + buffer_offset_ - buffer_aligned_offset_;
+  // Map into memory.
+  buffer_ = mmap(/*addr=*/nullptr, buffer_aligned_size_, PROT_READ, MAP_SHARED,
+                 fd, buffer_aligned_offset_);
+  if (buffer_ == MAP_FAILED) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown,
+        absl::StrFormat("Unable to map file to memory buffer, errno=%d", errno),
+        TfLiteSupportStatus::kFileMmapError);
+  }
+  return absl::OkStatus();
+}
+
+absl::string_view ExternalFileHandler::GetFileContent() {
+  if (!external_file_.file_content().empty()) {
+    return external_file_.file_content();
+  } else {
+    return absl::string_view(static_cast<const char*>(buffer_) +
+                                 buffer_offset_ - buffer_aligned_offset_,
+                             buffer_size_);
+  }
+}
+
+ExternalFileHandler::~ExternalFileHandler() {
+  if (buffer_ != MAP_FAILED) {
+    munmap(buffer_, buffer_aligned_size_);
+  }
+  if (owned_fd_ >= 0) {
+    close(owned_fd_);
+  }
+}
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/core/external_file_handler.h b/tensorflow_lite_support/cc/task/core/external_file_handler.h
new file mode 100644
index 0000000..236d903
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/external_file_handler.h
@@ -0,0 +1,94 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_EXTERNAL_FILE_HANDLER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_EXTERNAL_FILE_HANDLER_H_
+
+#include <memory>
+
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+
+namespace tflite {
+namespace task {
+namespace core {
+
+// Handler providing easy access to the contents of a file specified by an
+// ExternalFile proto [1]. Takes care (if needed, depending on the provided
+// proto fields) of opening and/or mapping the file in memory at creation time,
+// as well as closing and/or unmapping at destruction time.
+//
+// [1]: support/c/task/core/proto/external_file.proto
+class ExternalFileHandler {
+ public:
+  // Creates an ExternalFileHandler from the input ExternalFile proto and
+  // returns a pointer to the new object. Ownership is transferred to the
+  // caller. Returns an error if the creation failed, which may happen if the
+  // provided ExternalFile can't be opened or mapped into memory.
+  //
+  // Warning: Does not take ownership of `external_file`, which must refer to a
+  // valid proto that outlives this object.
+  static tflite::support::StatusOr<std::unique_ptr<ExternalFileHandler>>
+  CreateFromExternalFile(const ExternalFile* external_file);
+
+  ~ExternalFileHandler();
+
+  // Returns the content of the ExternalFile as a string_view guaranteed to be
+  // valid as long as the ExternalFileHandler is alive.
+  absl::string_view GetFileContent();
+
+ private:
+  // Private constructor, called from CreateFromExternalFile().
+  explicit ExternalFileHandler(const ExternalFile* external_file)
+      : external_file_(*external_file) {}
+
+  // Opens (if provided by path) and maps (if provided by path or file
+  // descriptor) the external file in memory. Does nothing otherwise, as file
+  // contents are already loaded in memory.
+  absl::Status MapExternalFile();
+
+  // Reference to the input ExternalFile.
+  const ExternalFile& external_file_;
+
+  // The file descriptor of the ExternalFile if provided by path, as it is
+  // opened and owned by this class. Set to -1 otherwise.
+  int owned_fd_{-1};
+
+  // Points to the memory buffer mapped from the file descriptor of the
+  // ExternalFile, if provided by path or file descriptor.
+  void* buffer_{};
+
+  // The mapped memory buffer offset, if any.
+  int64 buffer_offset_{};
+  // The size in bytes of the mapped memory buffer, if any.
+  int64 buffer_size_{};
+
+  // As mmap(2) requires the offset to be a multiple of sysconf(_SC_PAGE_SIZE):
+
+  // The aligned mapped memory buffer offset, if any.
+  int64 buffer_aligned_offset_{};
+  // The aligned mapped memory buffer size in bytes taking into account the
+  // offset shift introduced by buffer_aligned_memory_offset_, if any.
+  int64 buffer_aligned_size_{};
+};
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_EXTERNAL_FILE_HANDLER_H_
diff --git a/tensorflow_lite_support/cc/task/core/proto/BUILD b/tensorflow_lite_support/cc/task/core/proto/BUILD
new file mode 100644
index 0000000..7418e5b
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/proto/BUILD
@@ -0,0 +1,27 @@
+load("//tensorflow_lite_support/cc/port:build_defs.bzl", "support_cc_proto_library")
+
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+proto_library(
+    name = "external_file_proto",
+    srcs = ["external_file.proto"],
+)
+
+support_cc_proto_library(
+    name = "external_file_cc_proto",
+    srcs = ["external_file.proto"],
+    deps = [
+        ":external_file_proto",
+    ],
+)
+
+cc_library(
+    name = "external_file_proto_inc",
+    hdrs = ["external_file_proto_inc.h"],
+    deps = [":external_file_cc_proto"],
+)
diff --git a/tensorflow_lite_support/cc/task/core/proto/external_file.proto b/tensorflow_lite_support/cc/task/core/proto/external_file.proto
new file mode 100644
index 0000000..c0a4212
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/proto/external_file.proto
@@ -0,0 +1,67 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.core;
+
+
+// Represents external files used by the Task APIs (e.g. TF Lite FlatBuffer or
+// plain-text labels file). The files can be specified by one of the following
+// three ways:
+//
+// (1) file contents loaded in `file_content`.
+// (2) file path in `file_name`.
+// (3) file descriptor through `file_descriptor_meta` as returned by open(2).
+//
+// If more than one field of these fields is provided, they are used in this
+// precedence order.
+// Next id: 5
+message ExternalFile {
+  // The path to the file to open and mmap in memory
+  optional string file_name = 1;
+
+  // The file contents as a byte array.
+  optional bytes file_content = 2;
+
+  // The file descriptor to a file opened with open(2), with optional additional
+  // offset and length information.
+  optional FileDescriptorMeta file_descriptor_meta = 4;
+
+  // Deprecated field numbers.
+  reserved 3;
+}
+
+// A proto defining file descriptor metadata for mapping file into memory using
+// mmap(2).
+message FileDescriptorMeta {
+  // File descriptor as returned by open(2).
+  optional int32 fd = 1;
+
+  // Optional length of the mapped memory. If not specified, the actual file
+  // size is used at runtime.
+  //
+  // This is an advanced option, e.g. this can be used on Android to specify the
+  // length of a given asset obtained from AssetFileDescriptor#getLength().
+  optional int64 length = 2;
+
+  // Optional starting offset in the file referred to by the file descriptor
+  // `fd`.
+  //
+  // This is an advanced option, e.g. this can be used on Android to specify the
+  // offset of a given asset obtained from AssetFileDescriptor#getStartOffset().
+  optional int64 offset = 3;
+}
+
diff --git a/tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h b/tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h
new file mode 100644
index 0000000..017aa65
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h
@@ -0,0 +1,20 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_PROTO_EXTERNAL_FILE_PROTO_INC_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_PROTO_EXTERNAL_FILE_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/core/proto/external_file.pb.h"
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_PROTO_EXTERNAL_FILE_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/core/proto/proto_config.pbtxt b/tensorflow_lite_support/cc/task/core/proto/proto_config.pbtxt
new file mode 100644
index 0000000..dafb0fd
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/proto/proto_config.pbtxt
@@ -0,0 +1,2 @@
+allow_all: true
+optimize_mode: LITE_RUNTIME
diff --git a/tensorflow_lite_support/cc/task/core/task_api_factory.h b/tensorflow_lite_support/cc/task/core/task_api_factory.h
new file mode 100644
index 0000000..06c3a01
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/task_api_factory.h
@@ -0,0 +1,100 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TASK_API_FACTORY_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TASK_API_FACTORY_H_
+
+#include <memory>
+
+#include "absl/status/status.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow/lite/kernels/op_macros.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/base_task_api.h"
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+
+namespace tflite {
+namespace task {
+namespace core {
+template <typename T>
+using EnableIfBaseUntypedTaskApiSubclass = typename std::enable_if<
+    std::is_base_of<BaseUntypedTaskApi, T>::value>::type*;
+
+// Template creator for all subclasses of BaseTaskApi
+class TaskAPIFactory {
+ public:
+  TaskAPIFactory() = delete;
+
+  template <typename T, EnableIfBaseUntypedTaskApiSubclass<T> = nullptr>
+  static tflite::support::StatusOr<std::unique_ptr<T>> CreateFromBuffer(
+      const char* buffer_data, size_t buffer_size,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+      int num_threads = 1) {
+    auto engine = absl::make_unique<TfLiteEngine>(std::move(resolver));
+    RETURN_IF_ERROR(engine->BuildModelFromFlatBuffer(buffer_data, buffer_size));
+    return CreateFromTfLiteEngine<T>(std::move(engine), num_threads);
+  }
+
+  template <typename T, EnableIfBaseUntypedTaskApiSubclass<T> = nullptr>
+  static tflite::support::StatusOr<std::unique_ptr<T>> CreateFromFile(
+      const string& file_name,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+      int num_threads = 1) {
+    auto engine = absl::make_unique<TfLiteEngine>(std::move(resolver));
+    RETURN_IF_ERROR(engine->BuildModelFromFile(file_name));
+    return CreateFromTfLiteEngine<T>(std::move(engine), num_threads);
+  }
+
+  template <typename T, EnableIfBaseUntypedTaskApiSubclass<T> = nullptr>
+  static tflite::support::StatusOr<std::unique_ptr<T>> CreateFromFileDescriptor(
+      int file_descriptor,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+      int num_threads = 1) {
+    auto engine = absl::make_unique<TfLiteEngine>(std::move(resolver));
+    RETURN_IF_ERROR(engine->BuildModelFromFileDescriptor(file_descriptor));
+    return CreateFromTfLiteEngine<T>(std::move(engine), num_threads);
+  }
+
+  template <typename T, EnableIfBaseUntypedTaskApiSubclass<T> = nullptr>
+  static tflite::support::StatusOr<std::unique_ptr<T>>
+  CreateFromExternalFileProto(
+      const ExternalFile* external_file,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+      int num_threads = 1) {
+    auto engine = absl::make_unique<TfLiteEngine>(std::move(resolver));
+    RETURN_IF_ERROR(engine->BuildModelFromExternalFileProto(external_file));
+    return CreateFromTfLiteEngine<T>(std::move(engine), num_threads);
+  }
+
+ private:
+  template <typename T, EnableIfBaseUntypedTaskApiSubclass<T> = nullptr>
+  static tflite::support::StatusOr<std::unique_ptr<T>> CreateFromTfLiteEngine(
+      std::unique_ptr<TfLiteEngine> engine, int num_threads) {
+    RETURN_IF_ERROR(engine->InitInterpreter(num_threads));
+    return absl::make_unique<T>(std::move(engine));
+  }
+};
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TASK_API_FACTORY_H_
diff --git a/tensorflow_lite_support/cc/task/core/task_utils.cc b/tensorflow_lite_support/cc/task/core/task_utils.cc
new file mode 100644
index 0000000..de733ae
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/task_utils.cc
@@ -0,0 +1,66 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+
+#include <fstream>
+
+#include "absl/strings/str_cat.h"
+
+namespace tflite {
+namespace task {
+namespace core {
+
+double Dequantize(const TfLiteTensor& tensor, int index) {
+  int32_t quantized_value = 0;
+  switch (tensor.type) {
+    case kTfLiteUInt8:
+      quantized_value = GetTensorData<uint8_t>(&tensor)[index];
+      break;
+    case kTfLiteInt8:
+      quantized_value = GetTensorData<int8_t>(&tensor)[index];
+      break;
+    case kTfLiteInt16:
+      quantized_value = GetTensorData<int16_t>(&tensor)[index];
+      break;
+    default:
+      TF_LITE_FATAL(
+          absl::StrCat(
+              "Invalid tensor type for dequantization ", tensor.name,
+              ". Requested kTfLiteUInt8, kTfLiteInt8 or kTfLiteInt16, got ",
+              TfLiteTypeGetName(tensor.type), ".")
+              .c_str());
+  }
+  return tensor.params.scale * (quantized_value - tensor.params.zero_point);
+}
+
+std::string GetStringAtIndex(const TfLiteTensor* labels, int index) {
+  const auto& strref = tflite::GetString(labels, index);
+  return std::string(strref.str, strref.len);
+}
+
+std::string LoadBinaryContent(const char* filename) {
+  std::ifstream input_file(filename, std::ios::binary | std::ios::ate);
+  // Find buffer size from input file, and load the buffer.
+  size_t buffer_size = input_file.tellg();
+  std::string buffer(buffer_size, '\0');
+  input_file.seekg(0, std::ios::beg);
+  input_file.read(const_cast<char*>(buffer.c_str()), buffer_size);
+  return buffer;
+}
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/core/task_utils.h b/tensorflow_lite_support/cc/task/core/task_utils.h
new file mode 100644
index 0000000..c1c3fc3
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/task_utils.h
@@ -0,0 +1,182 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TASK_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TASK_UTILS_H_
+
+#include <algorithm>
+#include <cstring>
+#include <numeric>
+#include <vector>
+
+#include "absl/memory/memory.h"
+#include "absl/strings/str_cat.h"
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
+#include "tensorflow/lite/kernels/op_macros.h"
+#include "tensorflow/lite/string_util.h"
+#include "tensorflow/lite/type_to_tflitetype.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace core {
+
+// Checks if data type of tensor is T and returns the pointer casted to T if
+// applicable, returns nullptr if tensor type is not T.
+// See type_to_tflitetype.h for a mapping from plain C++ type to TfLiteType.
+template <typename T>
+T* TypedTensor(const TfLiteTensor* tensor_ptr) {
+  if (tensor_ptr->type == typeToTfLiteType<T>()) {
+    return reinterpret_cast<T*>(tensor_ptr->data.raw);
+  }
+  return nullptr;
+}
+
+// Checks and returns type of a tensor, fails if tensor type is not T.
+template <typename T>
+T* AssertAndReturnTypedTensor(const TfLiteTensor* tensor) {
+  if (T* v = TypedTensor<T>(tensor)) return v;
+  // TODO(b/150903834): throw exceptions instead
+  TF_LITE_ASSERT(tensor->data.raw);
+  TF_LITE_FATAL(absl::StrCat("Type mismatch for tensor ", tensor->name,
+                             ". Requested ",
+                             TfLiteTypeGetName(typeToTfLiteType<T>()), ", got ",
+                             TfLiteTypeGetName(tensor->type), ".")
+                    .c_str());
+}
+
+// Populates tensor with array of data, fails if data type doesn't match tensor
+// type or has not the same number of elements.
+template <typename T>
+inline void PopulateTensor(const T* data, int num_elements,
+                           TfLiteTensor* tensor) {
+  T* v = AssertAndReturnTypedTensor<T>(tensor);
+  size_t bytes = num_elements * sizeof(T);
+  // TODO(b/150903834): throw exceptions instead
+  TF_LITE_ASSERT(tensor->bytes == bytes);
+  memcpy(v, data, bytes);
+}
+
+// Populates tensor with vector of data, fails if data type doesn't match tensor
+// type or has not the same number of elements.
+template <typename T>
+inline void PopulateTensor(const std::vector<T>& data, TfLiteTensor* tensor) {
+  return PopulateTensor<T>(data.data(), data.size(), tensor);
+}
+
+template <>
+inline void PopulateTensor<std::string>(const std::vector<std::string>& data,
+                                        TfLiteTensor* tensor) {
+  if (tensor->type != kTfLiteString) {
+    TF_LITE_FATAL(absl::StrCat("Type mismatch for tensor ", tensor->name,
+                               ". Requested STRING, got ",
+                               TfLiteTypeGetName(tensor->type), ".")
+                      .c_str());
+  }
+  tflite::DynamicBuffer input_buf;
+  for (const auto& value : data) {
+    input_buf.AddString(value.data(), value.length());
+  }
+  input_buf.WriteToTensorAsVector(tensor);
+}
+
+// Populates tensor one data item, fails if data type doesn't match tensor
+// type.
+template <typename T>
+inline void PopulateTensor(const T& data, TfLiteTensor* tensor) {
+  T* v = AssertAndReturnTypedTensor<T>(tensor);
+  *v = data;
+}
+
+template <>
+inline void PopulateTensor<std::string>(const std::string& data,
+                                        TfLiteTensor* tensor) {
+  tflite::DynamicBuffer input_buf;
+  input_buf.AddString(data.data(), data.length());
+  input_buf.WriteToTensorAsVector(tensor);
+}
+
+// Populates a vector from the tensor, fails if data type doesn't match tensor
+// type.
+template <typename T>
+inline void PopulateVector(const TfLiteTensor* tensor, std::vector<T>* data) {
+  AssertAndReturnTypedTensor<T>(tensor);
+  const T* results = GetTensorData<T>(tensor);
+  size_t num = tensor->bytes / sizeof(tensor->type);
+  data->reserve(num);
+  for (int i = 0; i < num; i++) {
+    data->emplace_back(results[i]);
+  }
+}
+
+template <>
+inline void PopulateVector<std::string>(const TfLiteTensor* tensor,
+                                        std::vector<std::string>* data) {
+  AssertAndReturnTypedTensor<std::string>(tensor);
+  int num = GetStringCount(tensor);
+  data->reserve(num);
+  for (int i = 0; i < num; i++) {
+    const auto& strref = tflite::GetString(tensor, i);
+    data->emplace_back(strref.str, strref.len);
+  }
+}
+
+// Returns the reversely sorted indices of a vector.
+template <typename T>
+std::vector<size_t> ReverseSortIndices(const std::vector<T>& v) {
+  std::vector<size_t> idx(v.size());
+  std::iota(idx.begin(), idx.end(), 0);
+
+  std::stable_sort(idx.begin(), idx.end(),
+                   [&v](size_t i1, size_t i2) { return v[i2] < v[i1]; });
+
+  return idx;
+}
+
+// Returns the original (dequantized) value of the 'index'-th element of
+// 'tensor.
+double Dequantize(const TfLiteTensor& tensor, int index);
+
+// Returns the index-th string from the tensor.
+std::string GetStringAtIndex(const TfLiteTensor* labels, int index);
+
+// Loads binary content of a file into a string.
+std::string LoadBinaryContent(const char* filename);
+
+// Gets the tensor from a vector of tensors with name specified inside metadata.
+template <typename TensorType>
+static TensorType* FindTensorByName(
+    const std::vector<TensorType*>& tensors,
+    const flatbuffers::Vector<flatbuffers::Offset<TensorMetadata>>*
+        tensor_metadatas,
+    const std::string& name) {
+  if (tensor_metadatas == nullptr ||
+      tensor_metadatas->size() != tensors.size()) {
+    return nullptr;
+  }
+  for (int i = 0; i < tensor_metadatas->size(); i++) {
+    if (strcmp(name.data(), tensor_metadatas->Get(i)->name()->c_str()) == 0) {
+      return tensors[i];
+    }
+  }
+  return nullptr;
+}
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TASK_UTILS_H_
diff --git a/tensorflow_lite_support/cc/task/core/tflite_engine.cc b/tensorflow_lite_support/cc/task/core/tflite_engine.cc
new file mode 100644
index 0000000..cf923f6
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/tflite_engine.cc
@@ -0,0 +1,297 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+
+#include <unistd.h>
+
+#include "absl/strings/match.h"
+#include "absl/strings/str_cat.h"
+#include "tensorflow/lite/builtin_ops.h"
+#include "tensorflow/lite/stderr_reporter.h"
+#include "tensorflow/lite/tools/verifier.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+
+#if TFLITE_USE_C_API
+#include "tensorflow/lite/c/c_api_experimental.h"
+#else
+#include "tensorflow/lite/kernels/register.h"
+#endif
+
+namespace tflite {
+namespace task {
+namespace core {
+
+#ifdef __ANDROID__
+// https://github.com/opencv/opencv/issues/14906
+// "ios_base::Init" object is not a part of Android's "iostream" header (in case
+// of clang toolchain, NDK 20).
+//
+// Ref1:
+// https://en.cppreference.com/w/cpp/io/ios_base/Init
+//       The header <iostream> behaves as if it defines (directly or indirectly)
+//       an instance of std::ios_base::Init with static storage duration
+//
+// Ref2:
+// https://github.com/gcc-mirror/gcc/blob/gcc-8-branch/libstdc%2B%2B-v3/include/std/iostream#L73-L74
+static std::ios_base::Init s_iostream_initializer;
+#endif
+
+using ::absl::StatusCode;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::TfLiteSupportStatus;
+
+int TfLiteEngine::ErrorReporter::Report(const char* format, va_list args) {
+  return std::vsnprintf(error_message, sizeof(error_message), format, args);
+}
+
+bool TfLiteEngine::Verifier::Verify(const char* data, int length,
+                                    tflite::ErrorReporter* reporter) {
+  return tflite::Verify(data, length, *op_resolver_, reporter);
+}
+
+#if TFLITE_USE_C_API
+TfLiteEngine::TfLiteEngine(std::unique_ptr<tflite::OpResolver> resolver)
+    : model_(nullptr, TfLiteModelDelete),
+      resolver_(std::move(resolver)),
+      verifier_(resolver_.get()) {}
+#else
+TfLiteEngine::TfLiteEngine(std::unique_ptr<tflite::OpResolver> resolver)
+    : model_(), resolver_(std::move(resolver)), verifier_(resolver_.get()) {}
+#endif
+
+std::vector<TfLiteTensor*> TfLiteEngine::GetInputs() {
+  Interpreter* interpreter = this->interpreter();
+  std::vector<TfLiteTensor*> tensors;
+  int input_count = InputCount(interpreter);
+  tensors.reserve(input_count);
+  for (int index = 0; index < input_count; index++) {
+    tensors.push_back(GetInput(interpreter, index));
+  }
+  return tensors;
+}
+
+std::vector<const TfLiteTensor*> TfLiteEngine::GetOutputs() {
+  Interpreter* interpreter = this->interpreter();
+  std::vector<const TfLiteTensor*> tensors;
+  int output_count = OutputCount(interpreter);
+  tensors.reserve(output_count);
+  for (int index = 0; index < output_count; index++) {
+    tensors.push_back(GetOutput(interpreter, index));
+  }
+  return tensors;
+}
+
+// The following function is adapted from the code in
+// tflite::FlatBufferModel::VerifyAndBuildFromBuffer.
+void TfLiteEngine::VerifyAndBuildModelFromBuffer(const char* buffer_data,
+                                                 size_t buffer_size) {
+#if TFLITE_USE_C_API
+  // First verify with the base flatbuffers verifier.
+  // This verifies that the model is a valid flatbuffer model.
+  flatbuffers::Verifier base_verifier(
+      reinterpret_cast<const uint8_t*>(buffer_data), buffer_size);
+  if (!VerifyModelBuffer(base_verifier)) {
+    TF_LITE_REPORT_ERROR(&error_reporter_,
+                         "The model is not a valid Flatbuffer buffer");
+    model_ = nullptr;
+    return;
+  }
+  // Next verify with the extra verifier.  This verifies that the model only
+  // uses operators supported by the OpResolver.
+  if (!verifier_.Verify(buffer_data, buffer_size, &error_reporter_)) {
+    model_ = nullptr;
+    return;
+  }
+  // Build the model.
+  model_.reset(TfLiteModelCreate(buffer_data, buffer_size));
+#else
+  model_ = tflite::FlatBufferModel::VerifyAndBuildFromBuffer(
+      buffer_data, buffer_size, &verifier_, &error_reporter_);
+#endif
+}
+
+absl::Status TfLiteEngine::InitializeFromModelFileHandler() {
+  const char* buffer_data = model_file_handler_->GetFileContent().data();
+  size_t buffer_size = model_file_handler_->GetFileContent().size();
+  VerifyAndBuildModelFromBuffer(buffer_data, buffer_size);
+  if (model_ == nullptr) {
+    // To be replaced with a proper switch-case when TF Lite model builder
+    // returns a `TfLiteStatus` code capturing this type of error.
+    if (absl::StrContains(error_reporter_.error_message,
+                          "The model is not a valid Flatbuffer")) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument, error_reporter_.error_message,
+          TfLiteSupportStatus::kInvalidFlatBufferError);
+    } else {
+      // TODO(b/154917059): augment status with another `TfLiteStatus` code when
+      // ready. And use a new `TfLiteStatus::kCoreTfLiteError` for the TFLS
+      // code, instead of the unspecified `kError`.
+      return CreateStatusWithPayload(
+          StatusCode::kUnknown,
+          absl::StrCat(
+              "Could not build model from the provided pre-loaded flatbuffer: ",
+              error_reporter_.error_message));
+    }
+  }
+
+  ASSIGN_OR_RETURN(
+      model_metadata_extractor_,
+      tflite::metadata::ModelMetadataExtractor::CreateFromModelBuffer(
+          buffer_data, buffer_size));
+
+  return absl::OkStatus();
+}
+
+absl::Status TfLiteEngine::BuildModelFromFlatBuffer(const char* buffer_data,
+                                                    size_t buffer_size) {
+  if (model_) {
+    return CreateStatusWithPayload(StatusCode::kInternal,
+                                   "Model already built");
+  }
+  external_file_.set_file_content(std::string(buffer_data, buffer_size));
+  ASSIGN_OR_RETURN(
+      model_file_handler_,
+      ExternalFileHandler::CreateFromExternalFile(&external_file_));
+  return InitializeFromModelFileHandler();
+}
+
+absl::Status TfLiteEngine::BuildModelFromFile(const std::string& file_name) {
+  if (model_) {
+    return CreateStatusWithPayload(StatusCode::kInternal,
+                                   "Model already built");
+  }
+  external_file_.set_file_name(file_name);
+  ASSIGN_OR_RETURN(
+      model_file_handler_,
+      ExternalFileHandler::CreateFromExternalFile(&external_file_));
+  return InitializeFromModelFileHandler();
+}
+
+absl::Status TfLiteEngine::BuildModelFromFileDescriptor(int file_descriptor) {
+  if (model_) {
+    return CreateStatusWithPayload(StatusCode::kInternal,
+                                   "Model already built");
+  }
+  external_file_.mutable_file_descriptor_meta()->set_fd(file_descriptor);
+  ASSIGN_OR_RETURN(
+      model_file_handler_,
+      ExternalFileHandler::CreateFromExternalFile(&external_file_));
+  return InitializeFromModelFileHandler();
+}
+
+absl::Status TfLiteEngine::BuildModelFromExternalFileProto(
+    const ExternalFile* external_file) {
+  if (model_) {
+    return CreateStatusWithPayload(StatusCode::kInternal,
+                                   "Model already built");
+  }
+  ASSIGN_OR_RETURN(model_file_handler_,
+                   ExternalFileHandler::CreateFromExternalFile(external_file));
+  return InitializeFromModelFileHandler();
+}
+
+absl::Status TfLiteEngine::InitInterpreter(int num_threads) {
+  tflite::proto::ComputeSettings compute_settings;
+  return InitInterpreter(compute_settings, num_threads);
+}
+
+#if TFLITE_USE_C_API
+const TfLiteRegistration* FindBuiltinOp(void* user_data,
+                                        TfLiteBuiltinOperator builtin_op,
+                                        int version) {
+  OpResolver* op_resolver = reinterpret_cast<OpResolver*>(user_data);
+  tflite::BuiltinOperator op = static_cast<tflite::BuiltinOperator>(builtin_op);
+  return op_resolver->FindOp(op, version);
+}
+
+const TfLiteRegistration* FindCustomOp(void* user_data, const char* custom_op,
+                                       int version) {
+  OpResolver* op_resolver = reinterpret_cast<OpResolver*>(user_data);
+  return op_resolver->FindOp(custom_op, version);
+}
+#endif
+
+absl::Status TfLiteEngine::InitInterpreter(
+    const tflite::proto::ComputeSettings& compute_settings, int num_threads) {
+  if (model_ == nullptr) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        "TF Lite FlatBufferModel is null. Please make sure to call one of the "
+        "BuildModelFrom methods before calling InitInterpreter.");
+  }
+#if TFLITE_USE_C_API
+  std::function<absl::Status(TfLiteDelegate*,
+                             std::unique_ptr<Interpreter, InterpreterDeleter>*)>
+      initializer = [this, num_threads](
+          TfLiteDelegate* optional_delegate,
+          std::unique_ptr<Interpreter, InterpreterDeleter>* interpreter_out)
+      -> absl::Status {
+    std::unique_ptr<TfLiteInterpreterOptions,
+                    void (*)(TfLiteInterpreterOptions*)>
+        options{TfLiteInterpreterOptionsCreate(),
+                TfLiteInterpreterOptionsDelete};
+    TfLiteInterpreterOptionsSetOpResolver(options.get(), FindBuiltinOp,
+                                          FindCustomOp, resolver_.get());
+    TfLiteInterpreterOptionsSetNumThreads(options.get(), num_threads);
+    if (optional_delegate != nullptr) {
+      TfLiteInterpreterOptionsAddDelegate(options.get(), optional_delegate);
+    }
+    interpreter_out->reset(
+        TfLiteInterpreterCreateWithSelectedOps(model_.get(), options.get()));
+    if (*interpreter_out == nullptr) {
+      return CreateStatusWithPayload(
+          StatusCode::kAborted,
+          absl::StrCat("Could not build the TF Lite interpreter: "
+                       "TfLiteInterpreterCreateWithSelectedOps failed: ",
+                       error_reporter_.error_message));
+    }
+    return absl::OkStatus();
+  };
+#else
+  auto initializer =
+      [this, num_threads](
+          std::unique_ptr<Interpreter, InterpreterDeleter>* interpreter_out)
+      -> absl::Status {
+    if (tflite::InterpreterBuilder(*model_, *resolver_)(
+            interpreter_out, num_threads) != kTfLiteOk) {
+      return CreateStatusWithPayload(
+          StatusCode::kUnknown,
+          absl::StrCat("Could not build the TF Lite interpreter: ",
+                       error_reporter_.error_message));
+    }
+    if (*interpreter_out == nullptr) {
+      return CreateStatusWithPayload(StatusCode::kInternal,
+                                     "TF Lite interpreter is null.");
+    }
+    return absl::OkStatus();
+  };
+#endif
+
+  absl::Status status =
+      interpreter_.InitializeWithFallback(initializer, compute_settings);
+
+  if (!status.ok() &&
+      !status.GetPayload(tflite::support::kTfLiteSupportPayload).has_value()) {
+    status = CreateStatusWithPayload(status.code(), status.message());
+  }
+  return status;
+}
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/core/tflite_engine.h b/tensorflow_lite_support/cc/task/core/tflite_engine.h
new file mode 100644
index 0000000..30f239d
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/core/tflite_engine.h
@@ -0,0 +1,245 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TFLITE_ENGINE_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TFLITE_ENGINE_H_
+
+#include <sys/mman.h>
+
+#include <memory>
+
+#include "absl/memory/memory.h"
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow/lite/kernels/register.h"
+#include "tensorflow_lite_support/cc/port/tflite_wrapper.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+
+// If compiled with -DTFLITE_USE_C_API, this file will use the TF Lite C API
+// rather than the TF Lite C++ API.
+// TODO(b/168025296): eliminate the '#if TFLITE_USE_C_API' directives here and
+// elsewhere and instead use the C API unconditionally, once we have a suitable
+// replacement for the features of tflite::support::TfLiteInterpreterWrapper.
+#if TFLITE_USE_C_API
+#include "tensorflow/lite/c/c_api.h"
+#include "tensorflow/lite/core/api/verifier.h"
+#include "tensorflow/lite/tools/verifier.h"
+#else
+#include "tensorflow/lite/interpreter.h"
+#include "tensorflow/lite/model.h"
+#endif
+
+namespace tflite {
+namespace task {
+namespace core {
+
+// TfLiteEngine encapsulates logic for TFLite model initialization, inference
+// and error reporting.
+class TfLiteEngine {
+ public:
+  // Types.
+  using InterpreterWrapper = tflite::support::TfLiteInterpreterWrapper;
+#if TFLITE_USE_C_API
+  using Model = struct TfLiteModel;
+  using Interpreter = struct TfLiteInterpreter;
+  using ModelDeleter = void (*)(Model*);
+  using InterpreterDeleter = InterpreterWrapper::InterpreterDeleter;
+#else
+  using Model = tflite::FlatBufferModel;
+  using Interpreter = tflite::Interpreter;
+  using ModelDeleter = std::default_delete<Model>;
+  using InterpreterDeleter = std::default_delete<Interpreter>;
+#endif
+
+  // Constructors.
+  explicit TfLiteEngine(
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+  // Model is neither copyable nor movable.
+  TfLiteEngine(const TfLiteEngine&) = delete;
+  TfLiteEngine& operator=(const TfLiteEngine&) = delete;
+
+  // Accessors.
+  static int32_t InputCount(const Interpreter* interpreter) {
+#if TFLITE_USE_C_API
+    return TfLiteInterpreterGetInputTensorCount(interpreter);
+#else
+    return interpreter->inputs().size();
+#endif
+  }
+  static int32_t OutputCount(const Interpreter* interpreter) {
+#if TFLITE_USE_C_API
+    return TfLiteInterpreterGetOutputTensorCount(interpreter);
+#else
+    return interpreter->outputs().size();
+#endif
+  }
+  static TfLiteTensor* GetInput(Interpreter* interpreter, int index) {
+#if TFLITE_USE_C_API
+    return TfLiteInterpreterGetInputTensor(interpreter, index);
+#else
+    return interpreter->tensor(interpreter->inputs()[index]);
+#endif
+  }
+  // Same as above, but const.
+  static const TfLiteTensor* GetInput(const Interpreter* interpreter,
+                                      int index) {
+#if TFLITE_USE_C_API
+    return TfLiteInterpreterGetInputTensor(interpreter, index);
+#else
+    return interpreter->tensor(interpreter->inputs()[index]);
+#endif
+  }
+  static TfLiteTensor* GetOutput(Interpreter* interpreter, int index) {
+#if TFLITE_USE_C_API
+    // We need a const_cast here, because the TF Lite C API only has a non-const
+    // version of GetOutputTensor (in part because C doesn't support overloading
+    // on const).
+    return const_cast<TfLiteTensor*>(
+        TfLiteInterpreterGetOutputTensor(interpreter, index));
+#else
+    return interpreter->tensor(interpreter->outputs()[index]);
+#endif
+  }
+  // Same as above, but const.
+  static const TfLiteTensor* GetOutput(const Interpreter* interpreter,
+                                       int index) {
+#if TFLITE_USE_C_API
+    return TfLiteInterpreterGetOutputTensor(interpreter, index);
+#else
+    return interpreter->tensor(interpreter->outputs()[index]);
+#endif
+  }
+
+  std::vector<TfLiteTensor*> GetInputs();
+  std::vector<const TfLiteTensor*> GetOutputs();
+
+  const Model* model() const { return model_.get(); }
+  Interpreter* interpreter() { return interpreter_.get(); }
+  const Interpreter* interpreter() const { return interpreter_.get(); }
+  InterpreterWrapper* interpreter_wrapper() { return &interpreter_; }
+  const tflite::metadata::ModelMetadataExtractor* metadata_extractor() const {
+    return model_metadata_extractor_.get();
+  }
+
+  // Builds the TF Lite FlatBufferModel (model_) from the raw FlatBuffer data
+  // whose ownership remains with the caller, and which must outlive the current
+  // object. This performs extra verification on the input data using
+  // tflite::Verify.
+  absl::Status BuildModelFromFlatBuffer(const char* buffer_data,
+                                        size_t buffer_size);
+
+  // Builds the TF Lite model from a given file.
+  absl::Status BuildModelFromFile(const std::string& file_name);
+
+  // Builds the TF Lite model from a given file descriptor using mmap(2).
+  absl::Status BuildModelFromFileDescriptor(int file_descriptor);
+
+  // Builds the TFLite model from the provided ExternalFile proto, which must
+  // outlive the current object.
+  absl::Status BuildModelFromExternalFileProto(
+      const ExternalFile* external_file);
+
+  // Initializes interpreter with encapsulated model.
+  // Note: setting num_threads to -1 has for effect to let TFLite runtime set
+  // the value.
+  absl::Status InitInterpreter(int num_threads = 1);
+
+  // Same as above, but allows specifying `compute_settings` for acceleration.
+  absl::Status InitInterpreter(
+      const tflite::proto::ComputeSettings& compute_settings,
+      int num_threads = 1);
+
+  // Cancels the on-going `Invoke()` call if any and if possible. This method
+  // can be called from a different thread than the one where `Invoke()` is
+  // running.
+  void Cancel() {
+#if TFLITE_USE_C_API
+    // NOP.
+#else
+    interpreter_.Cancel();
+#endif
+  }
+
+ protected:
+  // TF Lite's DefaultErrorReporter() outputs to stderr. This one captures the
+  // error into a string so that it can be used to complement tensorflow::Status
+  // error messages.
+  struct ErrorReporter : public tflite::ErrorReporter {
+    // Last error message captured by this error reporter.
+    char error_message[256];
+    int Report(const char* format, va_list args) override;
+  };
+  // Custom error reporter capturing low-level TF Lite error messages.
+  ErrorReporter error_reporter_;
+
+ private:
+  // Direct wrapper around tflite::TfLiteVerifier which checks the integrity of
+  // the FlatBuffer data provided as input.
+  class Verifier : public tflite::TfLiteVerifier {
+   public:
+    explicit Verifier(const tflite::OpResolver* op_resolver)
+        : op_resolver_(op_resolver) {}
+    bool Verify(const char* data, int length,
+                tflite::ErrorReporter* reporter) override;
+    // The OpResolver to be used to build the TF Lite interpreter.
+    const tflite::OpResolver* op_resolver_;
+  };
+
+  // Verifies that the supplied buffer refers to a valid flatbuffer model,
+  // and that it uses only operators that are supported by the OpResolver
+  // that was passed to the TfLiteEngine constructor, and then builds
+  // the model from the buffer and stores it in 'model_'.
+  void VerifyAndBuildModelFromBuffer(const char* buffer_data,
+                                     size_t buffer_size);
+
+  // Gets the buffer from the file handler; verifies and builds the model
+  // from the buffer; if successful, sets 'model_metadata_extractor_' to be
+  // a TF Lite Metadata extractor for the model; and calculates an appropriate
+  // return Status,
+  absl::Status InitializeFromModelFileHandler();
+
+  // TF Lite model and interpreter for actual inference.
+  std::unique_ptr<Model, ModelDeleter> model_;
+
+  // Interpreter wrapper built from the model.
+  InterpreterWrapper interpreter_;
+
+  // TFLite Metadata extractor built from the model.
+  std::unique_ptr<tflite::metadata::ModelMetadataExtractor>
+      model_metadata_extractor_;
+
+  // Mechanism used by TF Lite to map Ops referenced in the FlatBuffer model to
+  // actual implementation. Defaults to TF Lite BuiltinOpResolver.
+  std::unique_ptr<tflite::OpResolver> resolver_;
+
+  // Extra verifier for FlatBuffer input data.
+  Verifier verifier_;
+
+  // ExternalFile and corresponding ExternalFileHandler for models loaded from
+  // disk or file descriptor.
+  ExternalFile external_file_;
+  std::unique_ptr<ExternalFileHandler> model_file_handler_;
+};
+
+}  // namespace core
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_CORE_TFLITE_ENGINE_H_
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/BUILD b/tensorflow_lite_support/cc/task/text/nlclassifier/BUILD
new file mode 100644
index 0000000..33b6f6a
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/BUILD
@@ -0,0 +1,118 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "bert_nl_classifier_c_api.h",
+    "nl_classifier_c_api.h",
+    "nl_classifier_c_api_common.h",
+])
+
+cc_library(
+    name = "nl_classifier",
+    srcs = [
+        "nl_classifier.cc",
+    ],
+    hdrs = [
+        "nl_classifier.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:base_task_api",
+        "//tensorflow_lite_support/cc/task/core:category",
+        "//tensorflow_lite_support/cc/task/core:task_api_factory",
+        "//tensorflow_lite_support/cc/task/core:task_utils",
+        "//tensorflow_lite_support/cc/text/tokenizers:regex_tokenizer",
+        "//tensorflow_lite_support/cc/text/tokenizers:tokenizer",
+        "//tensorflow_lite_support/cc/utils:common_utils",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/algorithm:container",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:string",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+    ],
+)
+
+cc_library(
+    name = "nl_classifier_c_api",
+    srcs = [
+        "nl_classifier_c_api.cc",
+    ],
+    hdrs = [
+        "nl_classifier_c_api.h",
+        "nl_classifier_c_api_common.h",
+    ],
+    visibility = ["//tensorflow_lite_support:__subpackages__"],
+    deps = [
+        ":nl_classifier",
+        ":nl_classifier_c_api_common",
+        "//tensorflow_lite_support/cc/task/core:category",
+        "@com_google_absl//absl/strings",
+    ],
+)
+
+cc_library(
+    name = "bert_nl_classifier",
+    srcs = [
+        "bert_nl_classifier.cc",
+    ],
+    hdrs = [
+        "bert_nl_classifier.h",
+    ],
+    deps = [
+        ":nl_classifier",
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:category",
+        "//tensorflow_lite_support/cc/task/core:task_api_factory",
+        "//tensorflow_lite_support/cc/task/core:task_utils",
+        "//tensorflow_lite_support/cc/text/tokenizers:tokenizer",
+        "//tensorflow_lite_support/cc/text/tokenizers:tokenizer_utils",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@org_tensorflow//tensorflow/lite:string",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+    ],
+)
+
+cc_library(
+    name = "bert_nl_classifier_c_api",
+    srcs = [
+        "bert_nl_classifier_c_api.cc",
+    ],
+    hdrs = [
+        "bert_nl_classifier_c_api.h",
+        "nl_classifier_c_api_common.h",
+    ],
+    visibility = ["//tensorflow_lite_support:__subpackages__"],
+    deps = [
+        ":bert_nl_classifier",
+        ":nl_classifier_c_api_common",
+        "//tensorflow_lite_support/cc/task/core:category",
+        "@com_google_absl//absl/strings",
+    ],
+)
+
+cc_library(
+    name = "nl_classifier_c_api_common",
+    srcs = [
+        "nl_classifier_c_api_common.cc",
+    ],
+    hdrs = [
+        "nl_classifier_c_api_common.h",
+    ],
+    visibility = ["//tensorflow_lite_support:__subpackages__"],
+)
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.cc b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.cc
new file mode 100644
index 0000000..d689c9e
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.cc
@@ -0,0 +1,198 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.h"
+
+#include <stddef.h>
+
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/strings/ascii.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow/lite/string_type.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/task/core/task_api_factory.h"
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+using ::tflite::support::text::tokenizer::CreateTokenizerFromProcessUnit;
+using ::tflite::support::text::tokenizer::TokenizerResult;
+using ::tflite::task::core::FindTensorByName;
+using ::tflite::task::core::PopulateTensor;
+
+namespace {
+constexpr char kIdsTensorName[] = "ids";
+constexpr char kMaskTensorName[] = "mask";
+constexpr char kSegmentIdsTensorName[] = "segment_ids";
+constexpr char kScoreTensorName[] = "probability";
+constexpr char kClassificationToken[] = "[CLS]";
+constexpr char kSeparator[] = "[SEP]";
+constexpr int kTokenizerProcessUnitIndex = 0;
+}  // namespace
+
+absl::Status BertNLClassifier::Preprocess(
+    const std::vector<TfLiteTensor*>& input_tensors, const std::string& input) {
+  auto* input_tensor_metadatas =
+      GetMetadataExtractor()->GetInputTensorMetadata();
+  auto* ids_tensor =
+      FindTensorByName(input_tensors, input_tensor_metadatas, kIdsTensorName);
+  auto* mask_tensor =
+      FindTensorByName(input_tensors, input_tensor_metadatas, kMaskTensorName);
+  auto* segment_ids_tensor = FindTensorByName(
+      input_tensors, input_tensor_metadatas, kSegmentIdsTensorName);
+
+  std::string processed_input = input;
+  absl::AsciiStrToLower(&processed_input);
+
+  TokenizerResult input_tokenize_results;
+  input_tokenize_results = tokenizer_->Tokenize(processed_input);
+
+  // 2 accounts for [CLS], [SEP]
+  absl::Span<const std::string> query_tokens =
+      absl::MakeSpan(input_tokenize_results.subwords.data(),
+                     input_tokenize_results.subwords.data() +
+                         std::min(static_cast<size_t>(kMaxSeqLen - 2),
+                                  input_tokenize_results.subwords.size()));
+
+  std::vector<std::string> tokens;
+  tokens.reserve(2 + query_tokens.size());
+  // Start of generating the features.
+  tokens.push_back(kClassificationToken);
+  // For query input.
+  for (const auto& query_token : query_tokens) {
+    tokens.push_back(query_token);
+  }
+  // For Separation.
+  tokens.push_back(kSeparator);
+
+  std::vector<int> input_ids(kMaxSeqLen, 0);
+  std::vector<int> input_mask(kMaxSeqLen, 0);
+  // Convert tokens back into ids and set mask
+  for (int i = 0; i < tokens.size(); ++i) {
+    tokenizer_->LookupId(tokens[i], &input_ids[i]);
+    input_mask[i] = 1;
+  }
+  //                             |<-----------kMaxSeqLen---------->|
+  // input_ids                 [CLS] s1  s2...  sn [SEP]  0  0...  0
+  // input_masks                 1    1   1...  1    1    0  0...  0
+  // segment_ids                 0    0   0...  0    0    0  0...  0
+
+  PopulateTensor(input_ids, ids_tensor);
+  PopulateTensor(input_mask, mask_tensor);
+  PopulateTensor(std::vector<int>(kMaxSeqLen, 0), segment_ids_tensor);
+
+  return absl::OkStatus();
+}
+
+StatusOr<std::vector<core::Category>> BertNLClassifier::Postprocess(
+    const std::vector<const TfLiteTensor*>& output_tensors,
+    const std::string& /*input*/) {
+  if (output_tensors.size() != 1) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        absl::StrFormat("BertNLClassifier models are expected to have only 1 "
+                        "output, found %d",
+                        output_tensors.size()),
+        TfLiteSupportStatus::kInvalidNumOutputTensorsError);
+  }
+  const TfLiteTensor* scores = FindTensorByName(
+      output_tensors, GetMetadataExtractor()->GetOutputTensorMetadata(),
+      kScoreTensorName);
+
+  // optional labels extracted from metadata
+  return BuildResults(scores, /*labels=*/nullptr);
+}
+
+StatusOr<std::unique_ptr<BertNLClassifier>>
+BertNLClassifier::CreateFromFile(
+    const std::string& path_to_model_with_metadata,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  std::unique_ptr<BertNLClassifier> bert_nl_classifier;
+  ASSIGN_OR_RETURN(bert_nl_classifier,
+                   core::TaskAPIFactory::CreateFromFile<BertNLClassifier>(
+                       path_to_model_with_metadata, std::move(resolver)));
+  RETURN_IF_ERROR(bert_nl_classifier->InitializeFromMetadata());
+  return std::move(bert_nl_classifier);
+}
+
+StatusOr<std::unique_ptr<BertNLClassifier>>
+BertNLClassifier::CreateFromBuffer(
+    const char* model_with_metadata_buffer_data,
+    size_t model_with_metadata_buffer_size,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  std::unique_ptr<BertNLClassifier> bert_nl_classifier;
+  ASSIGN_OR_RETURN(bert_nl_classifier,
+                   core::TaskAPIFactory::CreateFromBuffer<BertNLClassifier>(
+                       model_with_metadata_buffer_data,
+                       model_with_metadata_buffer_size, std::move(resolver)));
+  RETURN_IF_ERROR(bert_nl_classifier->InitializeFromMetadata());
+  return std::move(bert_nl_classifier);
+}
+
+StatusOr<std::unique_ptr<BertNLClassifier>> BertNLClassifier::CreateFromFd(
+    int fd, std::unique_ptr<tflite::OpResolver> resolver) {
+  std::unique_ptr<BertNLClassifier> bert_nl_classifier;
+  ASSIGN_OR_RETURN(
+      bert_nl_classifier,
+      core::TaskAPIFactory::CreateFromFileDescriptor<BertNLClassifier>(
+          fd, std::move(resolver)));
+  RETURN_IF_ERROR(bert_nl_classifier->InitializeFromMetadata());
+  return std::move(bert_nl_classifier);
+}
+
+absl::Status BertNLClassifier::InitializeFromMetadata() {
+  // Set up mandatory tokenizer.
+  const ProcessUnit* tokenizer_process_unit =
+      GetMetadataExtractor()->GetInputProcessUnit(kTokenizerProcessUnitIndex);
+  if (tokenizer_process_unit == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "No input process unit found from metadata.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+  ASSIGN_OR_RETURN(tokenizer_,
+                   CreateTokenizerFromProcessUnit(tokenizer_process_unit,
+                                                  GetMetadataExtractor()));
+
+  // Set up optional label vector.
+  TrySetLabelFromMetadata(
+      GetMetadataExtractor()->GetOutputTensorMetadata(kOutputTensorIndex))
+      .IgnoreError();
+  return absl::OkStatus();
+}
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.h b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.h
new file mode 100644
index 0000000..0c709ee
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.h
@@ -0,0 +1,105 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_BERT_NL_CLASSIFIER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_BERT_NL_CLASSIFIER_H_
+
+#include <stddef.h>
+
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow/lite/kernels/register.h"
+#include "tensorflow/lite/string_type.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+// Classifier API for NLClassification tasks with Bert models, categorizes
+// string into different classes.
+//
+// The API expects a Bert based TFLite model with metadata populated.
+// The metadata should contain the following information:
+//   - input_process_units for Wordpiece/Sentencepiece Tokenizer
+//   - 3 input tensors with names "ids", "mask" and "segment_ids"
+//   - 1 output tensor of type float32[1, 2], with a optionally attached label
+//     file. If a label file is attached, the file should be a plain text file
+//     with one label per line, the number of labels should match the number of
+//     categories the model outputs.
+
+class BertNLClassifier : public NLClassifier {
+ public:
+  using NLClassifier::NLClassifier;
+  // Max number of tokens to pass to the model.
+  static constexpr int kMaxSeqLen = 128;
+
+  // Factory function to create a BertNLClassifier from TFLite model with
+  // metadata.
+  static tflite::support::StatusOr<std::unique_ptr<BertNLClassifier>>
+  CreateFromFile(
+      const std::string& path_to_model_with_metadata,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Factory function to create a BertNLClassifier from in memory buffer of a
+  // TFLite model with metadata.
+  static tflite::support::StatusOr<std::unique_ptr<BertNLClassifier>>
+  CreateFromBuffer(
+      const char* model_with_metadata_buffer_data,
+      size_t model_with_metadata_buffer_size,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Factory function to create a BertNLClassifier from the file descriptor of a
+  // TFLite model with metadata.
+  static tflite::support::StatusOr<std::unique_ptr<BertNLClassifier>>
+  CreateFromFd(
+      int fd, std::unique_ptr<tflite::OpResolver> resolver =
+                  absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+ protected:
+  // Run tokenization on input text and construct three input tensors ids, mask
+  // and segment_ids for the model input.
+  absl::Status Preprocess(const std::vector<TfLiteTensor*>& input_tensors,
+                          const std::string& input) override;
+
+  // Extract model output and create results with label file attached in
+  // metadata. If no label file is attached, use output score index as labels.
+  tflite::support::StatusOr<std::vector<core::Category>> Postprocess(
+      const std::vector<const TfLiteTensor*>& output_tensors,
+      const std::string& input) override;
+
+ private:
+  // Initialize the API with the tokenizer and label files set in the metadata.
+  absl::Status InitializeFromMetadata();
+
+  std::unique_ptr<tflite::support::text::tokenizer::Tokenizer> tokenizer_;
+};
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_BERT_NL_CLASSIFIER_H_
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.cc b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.cc
new file mode 100644
index 0000000..0decc49
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.cc
@@ -0,0 +1,70 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.h"
+
+#include <memory>
+
+#include "absl/strings/string_view.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.h"
+
+using CategoryCPP = ::tflite::task::core::Category;
+using BertNLClassifierCPP =
+    ::tflite::task::text::nlclassifier::BertNLClassifier;
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+struct BertNLClassifier {
+  std::unique_ptr<BertNLClassifierCPP> impl;
+};
+
+BertNLClassifier* BertNLClassifierFromFile(const char* model_path) {
+  auto classifier_status =
+      BertNLClassifierCPP::CreateFromFile(std::string(model_path));
+  if (classifier_status.ok()) {
+    return new BertNLClassifier{.impl = std::unique_ptr<BertNLClassifierCPP>(
+                                    dynamic_cast<BertNLClassifierCPP*>(
+                                        classifier_status.value().release()))};
+  } else {
+    return nullptr;
+  }
+}
+
+Categories* BertNLClassifierClassify(const BertNLClassifier* classifier,
+                                     const char* text) {
+  std::vector<CategoryCPP> results =
+      classifier->impl->Classify(absl::string_view(text).data());
+  size_t size = results.size();
+  auto* categories = new Category[size];
+
+  for (size_t i = 0; i < size; ++i) {
+    categories[i].text = strdup(results[i].class_name.c_str());
+    categories[i].score = results[i].score;
+  }
+
+  auto* c_categories = new Categories;
+  c_categories->size = size;
+  c_categories->categories = categories;
+  return c_categories;
+}
+
+void BertNLClassifierDelete(BertNLClassifier* classifier) { delete classifier; }
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.h b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.h
new file mode 100644
index 0000000..1d0b8b6
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.h
@@ -0,0 +1,60 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_BERT_NL_CLASSIFIER_C_API_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_BERT_NL_CLASSIFIER_C_API_H_
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h"
+// --------------------------------------------------------------------------
+/// C API for BertNLClassifier.
+///
+/// The API leans towards simplicity and uniformity instead of convenience, as
+/// most usage will be by language-specific wrappers. It provides largely the
+/// same set of functionality as that of the C++ TensorFlow Lite
+/// `BertNLClassifier` API, but is useful for shared libraries where having
+/// a stable ABI boundary is important.
+///
+/// Usage:
+/// <pre><code>
+/// // Create the model and interpreter options.
+/// BertNLClassifier* classifier =
+///   BertNLClassifierFromFile("/path/to/model.tflite");
+///
+/// // classification.
+/// Categories* categories = Classify(classifier, context, question);
+///
+/// // Dispose of the API object.
+/// BertNLClassifierrDelete(classifier);
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+typedef struct BertNLClassifier BertNLClassifier;
+
+// Creates BertNLClassifier from model path, returns nullptr if the file
+// doesn't exist or is not a well formatted TFLite model path.
+extern BertNLClassifier* BertNLClassifierFromFile(const char* model_path);
+
+// Invokes the encapsulated TFLite model and classifies the input text.
+extern struct Categories* BertNLClassifierClassify(
+    const BertNLClassifier* classifier, const char* text);
+
+extern void BertNLClassifierDelete(BertNLClassifier* classifier);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_BERT_NL_CLASSIFIER_C_API_H_
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.cc b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.cc
new file mode 100644
index 0000000..1643e3e
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.cc
@@ -0,0 +1,467 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h"
+
+#include <cstddef>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/algorithm/container.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_cat.h"
+#include "absl/strings/string_view.h"
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/task/core/task_api_factory.h"
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+#include "tensorflow_lite_support/cc/utils/common_utils.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+using ::absl::StatusCode;
+using ::flatbuffers::Offset;
+using ::flatbuffers::Vector;
+using ::tflite::TensorMetadata;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+using ::tflite::support::text::tokenizer::RegexTokenizer;
+using ::tflite::support::text::tokenizer::Tokenizer;
+using ::tflite::support::text::tokenizer::TokenizerResult;
+using ::tflite::support::utils::LoadVocabFromBuffer;
+using ::tflite::task::core::Category;
+using ::tflite::task::core::Dequantize;
+using ::tflite::task::core::GetStringAtIndex;
+using ::tflite::task::core::PopulateTensor;
+
+namespace {
+constexpr int kRegexTokenizerInputTensorIndex = 0;
+constexpr int kRegexTokenizerProcessUnitIndex = 0;
+
+StatusOr<absl::string_view> CheckAndLoadFirstAssociatedFile(
+    const flatbuffers::Vector<flatbuffers::Offset<tflite::AssociatedFile>>*
+        associated_files,
+    const tflite::metadata::ModelMetadataExtractor* metadata_extractor) {
+  if (associated_files == nullptr || associated_files->size() < 1 ||
+      associated_files->Get(0)->name() == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "Invalid vocab_file from input process unit.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+  ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
+                   metadata_extractor->GetAssociatedFile(
+                       associated_files->Get(0)->name()->str()));
+  return vocab_buffer;
+}
+
+StatusOr<std::unique_ptr<Tokenizer>> CreateRegexTokenizerFromProcessUnit(
+    const tflite::ProcessUnit* tokenizer_process_unit,
+    const tflite::metadata::ModelMetadataExtractor* metadata_extractor) {
+  if (metadata_extractor == nullptr || tokenizer_process_unit == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "No metadata or input process unit found.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+
+  if (tokenizer_process_unit->options_type() !=
+      ProcessUnitOptions_RegexTokenizerOptions) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kNotFound,
+        absl::StrCat(
+            "Incorrect options_type:", tokenizer_process_unit->options_type(),
+            " need RegexTokenizerOptions."),
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+
+  const tflite::RegexTokenizerOptions* options =
+      tokenizer_process_unit->options_as<RegexTokenizerOptions>();
+  ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
+                   CheckAndLoadFirstAssociatedFile(options->vocab_file(),
+                                                   metadata_extractor));
+  if (options->delim_regex_pattern() == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "Invalid delim_regex_pattern from input process unit.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+
+  std::unique_ptr<RegexTokenizer> regex_tokenizer =
+      absl::make_unique<RegexTokenizer>(options->delim_regex_pattern()->str(),
+                                        vocab_buffer.data(),
+                                        vocab_buffer.size());
+
+  int unknown_token_id = 0;
+  if (!regex_tokenizer->GetUnknownToken(&unknown_token_id)) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "RegexTokenizer doesn't have <UNKNOWN> token.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+
+  int pad_token_id = 0;
+  if (!regex_tokenizer->GetPadToken(&pad_token_id)) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "RegexTokenizer doesn't have <PAD> token.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+  return regex_tokenizer;
+}
+
+}  // namespace
+
+const NLClassifierOptions& NLClassifier::GetOptions() const { return options_; }
+
+absl::Status NLClassifier::TrySetLabelFromMetadata(
+    const TensorMetadata* metadata) {
+  if (metadata == nullptr) {
+    return CreateStatusWithPayload(absl::StatusCode::kInvalidArgument,
+                                   "Metadata not found for output tensor",
+                                   TfLiteSupportStatus::kMetadataNotFoundError);
+  }
+  const auto* associated_files = metadata->associated_files();
+  if (associated_files == nullptr || associated_files->size() == 0) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "No label file found for tensor metadata.",
+        TfLiteSupportStatus::kMetadataMissingLabelsError);
+  }
+  const tflite::AssociatedFile* associated_file =
+      associated_files->Get(kOutputTensorLabelFileIndex);
+  if (associated_file->type() != AssociatedFileType_TENSOR_AXIS_LABELS) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "Incorrect label type found for tensor metadata.",
+        TfLiteSupportStatus::kMetadataMissingLabelsError);
+  }
+  tflite::support::StatusOr<absl::string_view> label_buffer =
+      GetMetadataExtractor()->GetAssociatedFile(
+          associated_files->Get(kOutputTensorIndex)->name()->str());
+  if (label_buffer.ok()) {
+    labels_vector_ =
+        absl::make_unique<std::vector<std::string>>(LoadVocabFromBuffer(
+            label_buffer.value().data(), label_buffer.value().size()));
+    return absl::OkStatus();
+  } else {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "Failed to extract label file from metadata.",
+        TfLiteSupportStatus::kMetadataMissingLabelsError);
+  }
+}
+
+std::vector<Category> NLClassifier::Classify(const std::string& text) {
+  // The NLClassifier implementation for Preprocess() and Postprocess() never
+  // returns errors: just call value().
+  return Infer(text).value();
+}
+
+absl::Status NLClassifier::Preprocess(
+    const std::vector<TfLiteTensor*>& input_tensors, const std::string& input) {
+  TfLiteTensor* input_tensor = FindTensorWithNameOrIndex(
+      input_tensors, GetMetadataExtractor()->GetInputTensorMetadata(),
+      options_.input_tensor_name, options_.input_tensor_index);
+  if (input_tensor == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "No input tensor found from NLClassifierOptions.",
+        TfLiteSupportStatus::kInputTensorNotFoundError);
+  }
+
+  if (HasRegexTokenizerMetadata()) {
+    //                              |<-------sentence_length-------->|
+    // input_tensor                 <START>, t1, t2... <PAD>, <PAD>...
+    // <START> is optional, t1, t2... will be replaced by <UNKNOWN> if it's not
+    // found in tokenizer vocab.
+    TokenizerResult result = tokenizer_->Tokenize(input);
+
+    size_t max_sentence_length = input_tensor->dims->size == 2
+                                     ? input_tensor->dims->data[1]
+                                     : input_tensor->dims->data[0];
+
+    int unknown_token_id = 0;
+    tokenizer_->GetUnknownToken(&unknown_token_id);
+
+    int pad_token_id = 0;
+    tokenizer_->GetPadToken(&pad_token_id);
+
+    std::vector<int> input_tokens(max_sentence_length, pad_token_id);
+    int start_token_id = 0;
+    size_t input_token_index = 0;
+    if (tokenizer_->GetStartToken(&start_token_id)) {
+      input_tokens[0] = start_token_id;
+      input_token_index = 1;
+    }
+
+    for (size_t i = 0; (i < result.subwords.size()) &&
+                       (input_token_index < max_sentence_length);
+         ++i, ++input_token_index) {
+      const std::string& token = result.subwords[i];
+      int token_id = 0;
+      if (tokenizer_->LookupId(token, &token_id)) {
+        input_tokens[input_token_index] = token_id;
+      } else {
+        input_tokens[input_token_index] = unknown_token_id;
+      }
+    }
+
+    PopulateTensor(input_tokens, input_tensor);
+  } else {
+    PopulateTensor(input, input_tensor);
+  }
+  return absl::OkStatus();
+}
+
+StatusOr<std::vector<Category>> NLClassifier::Postprocess(
+    const std::vector<const TfLiteTensor*>& output_tensors,
+    const std::string& /*input*/) {
+  return BuildResults(
+      FindTensorWithNameOrIndex(
+          output_tensors, GetMetadataExtractor()->GetOutputTensorMetadata(),
+          options_.output_score_tensor_name,
+          options_.output_score_tensor_index),
+      FindTensorWithNameOrIndex(
+          output_tensors, GetMetadataExtractor()->GetInputTensorMetadata(),
+          options_.output_label_tensor_name,
+          options_.output_label_tensor_index));
+}
+
+std::vector<Category> NLClassifier::BuildResults(const TfLiteTensor* scores,
+                                                 const TfLiteTensor* labels) {
+  bool use_index_as_labels = (labels_vector_ == nullptr) && (labels == nullptr);
+  // Some models output scores with transposed shape [1, categories]
+  int categories =
+      scores->dims->size == 2 ? scores->dims->data[1] : scores->dims->data[0];
+
+  std::vector<Category> predictions;
+  predictions.reserve(categories);
+
+  bool should_dequantize = scores->type == kTfLiteUInt8 ||
+                           scores->type == kTfLiteInt8 ||
+                           scores->type == kTfLiteInt16;
+  for (int index = 0; index < categories; index++) {
+    std::string label;
+    if (use_index_as_labels) {
+      label = std::to_string(index);
+    } else if (labels_vector_ == nullptr) {
+      if (labels->type == kTfLiteString) {
+        label = GetStringAtIndex(labels, index);
+      } else if (labels->type == kTfLiteInt32) {
+        label = std::to_string(GetTensorData<int>(labels)[index]);
+      }
+    } else {
+      label = (*labels_vector_)[index];
+    }
+    if (should_dequantize) {
+      predictions.push_back(Category(label, Dequantize(*scores, index)));
+    } else if (scores->type == kTfLiteBool) {
+      predictions.push_back(
+          Category(label, GetTensorData<bool>(scores)[index] ? 1.0 : 0.0));
+    } else {
+      predictions.push_back(
+          Category(label, scores->type == kTfLiteFloat32
+                              ? GetTensorData<float>(scores)[index]
+                              : GetTensorData<double>(scores)[index]));
+    }
+  }
+
+  return predictions;
+}
+absl::Status NLClassifier::Initialize(const NLClassifierOptions& options) {
+  options_ = options;
+  // input tensor should be type STRING
+  auto input_tensor = FindTensorWithNameOrIndex(
+      GetInputTensors(), GetMetadataExtractor()->GetInputTensorMetadata(),
+      options.input_tensor_name, options.input_tensor_index);
+  if (input_tensor == nullptr) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat("No input tensor found with name ",
+                     options.input_tensor_name, " or at index ",
+                     options.input_tensor_index),
+        TfLiteSupportStatus::kInputTensorNotFoundError);
+  }
+  if (HasRegexTokenizerMetadata()) {
+    if (input_tensor->type != kTfLiteInt32) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrCat("Type mismatch for input tensor ", input_tensor->name,
+                       ". Requested INT32, got ",
+                       TfLiteTypeGetName(input_tensor->type), "."),
+          TfLiteSupportStatus::kInvalidInputTensorTypeError);
+    }
+    RETURN_IF_ERROR(SetupRegexTokenizer());
+  } else {
+    if (input_tensor->type != kTfLiteString) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrCat("Type mismatch for input tensor ", input_tensor->name,
+                       ". Requested STRING, got ",
+                       TfLiteTypeGetName(input_tensor->type), "."),
+          TfLiteSupportStatus::kInvalidInputTensorTypeError);
+    }
+  }
+
+  // output score tensor should be type
+  // UINT8/INT8/INT16(quantized) or FLOAT32/FLOAT64(dequantized) or BOOL
+  std::vector<const TfLiteTensor*> output_tensors = GetOutputTensors();
+  const Vector<Offset<TensorMetadata>>* output_tensor_metadatas =
+      GetMetadataExtractor()->GetOutputTensorMetadata();
+
+  const auto scores = FindTensorWithNameOrIndex(
+      output_tensors, output_tensor_metadatas, options.output_score_tensor_name,
+      options.output_score_tensor_index);
+  if (scores == nullptr) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat("No output score tensor found with name ",
+                     options.output_score_tensor_name, " or at index ",
+                     options.output_score_tensor_index),
+        TfLiteSupportStatus::kOutputTensorNotFoundError);
+  }
+  static constexpr TfLiteType valid_types[] = {kTfLiteUInt8,   kTfLiteInt8,
+                                               kTfLiteInt16,   kTfLiteFloat32,
+                                               kTfLiteFloat64, kTfLiteBool};
+  if (!absl::c_linear_search(valid_types, scores->type)) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat("Type mismatch for score tensor ", scores->name,
+                     ". Requested one of these types: "
+                     "INT8/UINT8/INT16/FLOAT32/FLOAT64/BOOL, got ",
+                     TfLiteTypeGetName(scores->type), "."),
+        TfLiteSupportStatus::kInvalidOutputTensorTypeError);
+  }
+
+  // Extract associated label file from output score tensor if one exists, a
+  // well-formatted metadata should have same number of tensors with the model.
+  if (output_tensor_metadatas &&
+      output_tensor_metadatas->size() == output_tensors.size()) {
+    for (int i = 0; i < output_tensor_metadatas->size(); ++i) {
+      const tflite::TensorMetadata* metadata = output_tensor_metadatas->Get(i);
+      if ((metadata->name() && metadata->name()->string_view() ==
+                                   options.output_score_tensor_name) ||
+          i == options.output_score_tensor_index) {
+        if (TrySetLabelFromMetadata(metadata).ok()) {
+          return absl::OkStatus();
+        }
+      }
+    }
+  }
+
+  // If labels_vector_ is not set up from metadata, try register output label
+  // tensor from options.
+  if (labels_vector_ == nullptr) {
+    // output label tensor should be type STRING or INT32 if the one exists
+    auto labels = FindTensorWithNameOrIndex(
+        output_tensors, output_tensor_metadatas,
+        options.output_label_tensor_name, options.output_label_tensor_index);
+    if (labels != nullptr && labels->type != kTfLiteString &&
+        labels->type != kTfLiteInt32) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrCat("Type mismatch for label tensor ", scores->name,
+                       ". Requested STRING or INT32, got ",
+                       TfLiteTypeGetName(scores->type), "."),
+          TfLiteSupportStatus::kInvalidOutputTensorTypeError);
+    }
+  }
+  return absl::OkStatus();
+}
+
+StatusOr<std::unique_ptr<NLClassifier>>
+NLClassifier::CreateFromBufferAndOptions(
+    const char* model_buffer_data, size_t model_buffer_size,
+    const NLClassifierOptions& options,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  std::unique_ptr<NLClassifier> nl_classifier;
+  ASSIGN_OR_RETURN(
+      nl_classifier,
+      core::TaskAPIFactory::CreateFromBuffer<NLClassifier>(
+          model_buffer_data, model_buffer_size, std::move(resolver)));
+  RETURN_IF_ERROR(nl_classifier->Initialize(options));
+  return std::move(nl_classifier);
+}
+
+StatusOr<std::unique_ptr<NLClassifier>> NLClassifier::CreateFromFileAndOptions(
+    const std::string& path_to_model, const NLClassifierOptions& options,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  std::unique_ptr<NLClassifier> nl_classifier;
+  ASSIGN_OR_RETURN(nl_classifier,
+                   core::TaskAPIFactory::CreateFromFile<NLClassifier>(
+                       path_to_model, std::move(resolver)));
+  RETURN_IF_ERROR(nl_classifier->Initialize(options));
+  return std::move(nl_classifier);
+}
+
+StatusOr<std::unique_ptr<NLClassifier>> NLClassifier::CreateFromFdAndOptions(
+    int fd, const NLClassifierOptions& options,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  std::unique_ptr<NLClassifier> nl_classifier;
+  ASSIGN_OR_RETURN(nl_classifier,
+                   core::TaskAPIFactory::CreateFromFileDescriptor<NLClassifier>(
+                       fd, std::move(resolver)));
+  RETURN_IF_ERROR(nl_classifier->Initialize(options));
+  return std::move(nl_classifier);
+}
+
+bool NLClassifier::HasRegexTokenizerMetadata() {
+  const TensorMetadata* input_tensor_metadata =
+      GetMetadataExtractor()->GetInputTensorMetadata(
+          kRegexTokenizerInputTensorIndex);
+  if (input_tensor_metadata == nullptr) {
+    return false;
+  }
+  tflite::support::StatusOr<const tflite::ProcessUnit*> status =
+      GetMetadataExtractor()->FindFirstProcessUnit(
+          *input_tensor_metadata, ProcessUnitOptions_RegexTokenizerOptions);
+  return status.ok() ? status.value() != nullptr : false;
+}
+
+absl::Status NLClassifier::SetupRegexTokenizer() {
+  ASSIGN_OR_RETURN(
+      std::unique_ptr<Tokenizer> base_tokenizer,
+      CreateRegexTokenizerFromProcessUnit(
+          GetMetadataExtractor()
+              ->GetInputTensorMetadata(kRegexTokenizerInputTensorIndex)
+              ->process_units()
+              ->Get(kRegexTokenizerProcessUnitIndex),
+          GetMetadataExtractor()));
+
+  tokenizer_ = std::unique_ptr<RegexTokenizer>(
+      dynamic_cast<RegexTokenizer*>(base_tokenizer.release()));
+
+  return absl::OkStatus();
+}
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h
new file mode 100644
index 0000000..2a9573a
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h
@@ -0,0 +1,181 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_H_
+
+#include <stddef.h>
+#include <string.h>
+
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow/lite/kernels/register.h"
+#include "tensorflow/lite/string_type.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/base_task_api.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+// Options to identify input and output tensors of the model
+struct NLClassifierOptions {
+  int input_tensor_index = 0;
+  int output_score_tensor_index = 0;
+  // By default there is no output label tensor. The label file can be attached
+  // to the output score tensor metadata.
+  int output_label_tensor_index = -1;
+  std::string input_tensor_name = "INPUT";
+  std::string output_score_tensor_name = "OUTPUT_SCORE";
+  std::string output_label_tensor_name = "OUTPUT_LABEL";
+};
+
+// Classifier API for NLClassification tasks, categorizes string into different
+// classes.
+//
+// The API expects a TFLite model with the following input/output tensor:
+// Input tensor:
+//   (kTfLiteString) - input of the model, accepts a string.
+//      or
+//   (kTfLiteInt32) - input of the model, accepts a tokenized
+//   indices of a string input. A RegexTokenizer needs to be set up in the input
+//   tensor's metadata.
+// Output score tensor:
+//   (kTfLiteUInt8/kTfLiteInt8/kTfLiteInt16/kTfLiteFloat32/
+//    kTfLiteFloat64/kTfLiteBool)
+//    - output scores for each class, if type is one of the Int types,
+//      dequantize it to double, if type is kTfLiteBool, convert the values to
+//      0.0 and 1.0 respectively
+//    - can have an optional associated file in metadata for labels, the file
+//      should be a plain text file with one label per line, the number of
+//      labels should match the number of categories the model outputs.
+// Output label tensor: optional
+//   (kTfLiteString/kTfLiteInt32)
+//    - output classname for each class, should be of the same length with
+//      scores. If this tensor is not present, the API uses score indices as
+//      classnames.
+//    - will be ignored if output score tensor already has an associated label
+//      file.
+//
+// By default the API tries to find the input/output tensors with default
+// configurations in NLClassifierOptions, with tensor name prioritized over
+// tensor index. The option is configurable for different TFLite models.
+class NLClassifier : public core::BaseTaskApi<std::vector<core::Category>,
+                                              const std::string&> {
+ public:
+  using BaseTaskApi::BaseTaskApi;
+
+  // Creates a NLClassifier from TFLite model buffer.
+  static tflite::support::StatusOr<std::unique_ptr<NLClassifier>>
+  CreateFromBufferAndOptions(
+      const char* model_buffer_data, size_t model_buffer_size,
+      const NLClassifierOptions& options = {},
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Creates a NLClassifier from TFLite model file.
+  static tflite::support::StatusOr<std::unique_ptr<NLClassifier>>
+  CreateFromFileAndOptions(
+      const std::string& path_to_model, const NLClassifierOptions& options = {},
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Creates a NLClassifier from TFLite model file descriptor.
+  static tflite::support::StatusOr<std::unique_ptr<NLClassifier>>
+  CreateFromFdAndOptions(
+      int fd, const NLClassifierOptions& options = {},
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Performs classification on a string input, returns classified results.
+  std::vector<core::Category> Classify(const std::string& text);
+
+ protected:
+  static constexpr int kOutputTensorIndex = 0;
+  static constexpr int kOutputTensorLabelFileIndex = 0;
+
+  absl::Status Initialize(const NLClassifierOptions& options);
+  const NLClassifierOptions& GetOptions() const;
+
+  // Try to extract attached label file from metadata and initialize
+  // labels_vector_, return error if metadata type is incorrect or no label file
+  // is attached in metadata.
+  absl::Status TrySetLabelFromMetadata(const TensorMetadata* metadata);
+
+  // Pass through the input text into model's input tensor.
+  absl::Status Preprocess(const std::vector<TfLiteTensor*>& input_tensors,
+                          const std::string& input) override;
+
+  // Extract model output and create results with output label tensor or label
+  // file attached in metadata. If no output label tensor or label file is
+  // found, use output score index as labels.
+  tflite::support::StatusOr<std::vector<core::Category>> Postprocess(
+      const std::vector<const TfLiteTensor*>& output_tensors,
+      const std::string& input) override;
+
+  std::vector<core::Category> BuildResults(const TfLiteTensor* scores,
+                                           const TfLiteTensor* labels);
+
+  // Gets the tensor from a vector of tensors by checking tensor name first and
+  // tensor index second, return nullptr if no tensor is found.
+  template <typename TensorType>
+  static TensorType* FindTensorWithNameOrIndex(
+      const std::vector<TensorType*>& tensors,
+      const flatbuffers::Vector<flatbuffers::Offset<TensorMetadata>>*
+          metadata_array,
+      const std::string& name, int index) {
+    if (metadata_array != nullptr && metadata_array->size() == tensors.size()) {
+      for (int i = 0; i < metadata_array->size(); i++) {
+        if (strcmp(name.data(), metadata_array->Get(i)->name()->c_str()) == 0) {
+          return tensors[i];
+        }
+      }
+    }
+
+    for (TensorType* tensor : tensors) {
+      if (tensor->name == name) {
+        return tensor;
+      }
+    }
+    return index >= 0 && index < tensors.size() ? tensors[index] : nullptr;
+  }
+
+ private:
+  bool HasRegexTokenizerMetadata();
+  absl::Status SetupRegexTokenizer();
+
+  NLClassifierOptions options_;
+  // labels vector initialized from output tensor's associated file, if one
+  // exists.
+  std::unique_ptr<std::vector<std::string>> labels_vector_;
+  std::unique_ptr<tflite::support::text::tokenizer::RegexTokenizer> tokenizer_;
+};
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_H_
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.cc b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.cc
new file mode 100644
index 0000000..3f7827d
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.cc
@@ -0,0 +1,89 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.h"
+
+#include <memory>
+
+#include "absl/strings/string_view.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h"
+
+using CategoryCPP = ::tflite::task::core::Category;
+using NLClassifierCPP = ::tflite::task::text::nlclassifier::NLClassifier;
+using NLClassifierOptionsCPP =
+    ::tflite::task::text::nlclassifier::NLClassifierOptions;
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+struct NLClassifier {
+  std::unique_ptr<NLClassifierCPP> impl;
+};
+
+NLClassifier* NLClassifierFromFileAndOptions(
+    const char* model_path, const NLClassifierOptions* options) {
+  auto classifier_status = NLClassifierCPP::CreateFromFileAndOptions(
+      std::string(model_path),
+      {
+          .input_tensor_index = options->input_tensor_index,
+          .output_score_tensor_index = options->output_score_tensor_index,
+          .output_label_tensor_index = options->output_label_tensor_index,
+          .input_tensor_name = !options->input_tensor_name
+                                   ? ""
+                                   : std::string(options->input_tensor_name),
+          .output_score_tensor_name =
+              !options->output_score_tensor_name
+                  ? ""
+                  : std::string(options->output_score_tensor_name),
+          .output_label_tensor_name =
+              !options->output_label_tensor_name
+                  ? ""
+                  : std::string(options->output_label_tensor_name),
+      });
+
+  if (classifier_status.ok()) {
+    return new NLClassifier{
+        .impl = std::unique_ptr<NLClassifierCPP>(dynamic_cast<NLClassifierCPP*>(
+            classifier_status.value().release()))};
+  } else {
+    return nullptr;
+  }
+}
+
+Categories* NLClassifierClassify(const NLClassifier* classifier,
+                                 const char* text) {
+  std::vector<CategoryCPP> results =
+      classifier->impl->Classify(absl::string_view(text).data());
+  size_t size = results.size();
+  auto* categories = new Category[size];
+
+  for (size_t i = 0; i < size; ++i) {
+    categories[i].text = strdup(results[i].class_name.c_str());
+    categories[i].score = results[i].score;
+  }
+
+  auto* c_categories = new Categories;
+  c_categories->size = size;
+  c_categories->categories = categories;
+  return c_categories;
+}
+
+void NLClassifierDelete(NLClassifier* classifier) { delete classifier; }
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.h b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.h
new file mode 100644
index 0000000..1af93f2
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.h
@@ -0,0 +1,72 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_C_API_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_C_API_H_
+
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h"
+// --------------------------------------------------------------------------
+/// C API for NLClassifier.
+///
+/// The API leans towards simplicity and uniformity instead of convenience, as
+/// most usage will be by language-specific wrappers. It provides largely the
+/// same set of functionality as that of the C++ TensorFlow Lite `NLClassifier`
+/// API, but is useful for shared libraries where having a stable ABI boundary
+/// is important.
+///
+/// Usage:
+/// <pre><code>
+/// // Create the model and interpreter options.
+/// NLClassifier* classifier = NLClassifierFromFileAndOptions(
+///   "/path/to/model.tflite");
+///
+/// // classification.
+/// Categories* categories = Classify(classifier, context, question);
+///
+/// // Dispose of the API object.
+/// NLClassifierDelete(classifier);
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+typedef struct NLClassifier NLClassifier;
+
+struct NLClassifierOptions {
+  int input_tensor_index;
+  int output_score_tensor_index;
+  int output_label_tensor_index;
+  const char* input_tensor_name;
+  const char* output_score_tensor_name;
+  const char* output_label_tensor_name;
+};
+
+// Creates NLClassifier from model path and options, returns nullptr if the file
+// doesn't exist or is not a well formatted TFLite model path.
+extern NLClassifier* NLClassifierFromFileAndOptions(
+    const char* model_path,
+    const struct NLClassifierOptions* options);
+
+// Invokes the encapsulated TFLite model and classifies the input text.
+extern struct Categories* NLClassifierClassify(const NLClassifier* classifier,
+                                               const char* text);
+
+extern void NLClassifierDelete(NLClassifier* classifier);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_C_API_H_
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.cc b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.cc
new file mode 100644
index 0000000..3beb658
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.cc
@@ -0,0 +1,30 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+void NLClassifierCategoriesDelete(Categories* categories) {
+  delete[] categories->categories;
+  delete categories;
+}
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
diff --git a/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h
new file mode 100644
index 0000000..663c873
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h
@@ -0,0 +1,43 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_C_API_COMMON_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_C_API_COMMON_H_
+
+// Common structs shared between NLClassifier APIs
+//
+/// // Dispose of the Categories object.
+/// NLClassifierCategoriesDelete(categories);
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+struct Category {
+  char* text;
+  double score;
+};
+
+struct Categories {
+  int size;
+  struct Category* categories;
+};
+
+extern void NLClassifierCategoriesDelete(struct Categories* categories);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_NLCLASSIFIER_NL_CLASSIFIER_C_API_COMMON_H_
diff --git a/tensorflow_lite_support/cc/task/text/qa/BUILD b/tensorflow_lite_support/cc/task/text/qa/BUILD
new file mode 100644
index 0000000..49ad5a1
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/qa/BUILD
@@ -0,0 +1,61 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "bert_qa_c_api.h",
+])
+
+cc_library(
+    name = "question_answerer",
+    hdrs = [
+        "question_answerer.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/task/core:base_task_api",
+        "//tensorflow_lite_support/cc/task/core:tflite_engine",
+    ],
+)
+
+cc_library(
+    name = "bert_question_answerer",
+    srcs = [
+        "bert_question_answerer.cc",
+    ],
+    hdrs = [
+        "bert_question_answerer.h",
+    ],
+    deps = [
+        ":question_answerer",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:base_task_api",
+        "//tensorflow_lite_support/cc/task/core:task_api_factory",
+        "//tensorflow_lite_support/cc/task/core:task_utils",
+        "//tensorflow_lite_support/cc/task/core:tflite_engine",
+        "//tensorflow_lite_support/cc/text/tokenizers:bert_tokenizer",
+        "//tensorflow_lite_support/cc/text/tokenizers:sentencepiece_tokenizer",
+        "//tensorflow_lite_support/cc/text/tokenizers:tokenizer",
+        "//tensorflow_lite_support/cc/text/tokenizers:tokenizer_utils",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "@com_google_absl//absl/container:flat_hash_map",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+    ],
+)
+
+cc_library(
+    name = "bert_qa_c_api",
+    srcs = [
+        "bert_qa_c_api.cc",
+    ],
+    hdrs = [
+        "bert_qa_c_api.h",
+    ],
+    visibility = ["//tensorflow_lite_support:__subpackages__"],
+    deps = [
+        ":bert_question_answerer",
+        ":question_answerer",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.cc b/tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.cc
new file mode 100644
index 0000000..5fafb59
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.cc
@@ -0,0 +1,79 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.h"
+
+#include <memory>
+
+#include "tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.h"
+#include "tensorflow_lite_support/cc/task/text/qa/question_answerer.h"
+
+using BertQuestionAnswererCPP = ::tflite::task::text::qa::BertQuestionAnswerer;
+using QaAnswerCPP = ::tflite::task::text::qa::QaAnswer;
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+struct BertQuestionAnswerer {
+  std::unique_ptr<BertQuestionAnswererCPP> impl;
+};
+
+BertQuestionAnswerer* BertQuestionAnswererFromFile(const char* model_path) {
+  auto bert_qa_status =
+      BertQuestionAnswererCPP::CreateFromFile(std::string(model_path));
+  if (bert_qa_status.ok()) {
+    return new BertQuestionAnswerer{
+        .impl = std::unique_ptr<BertQuestionAnswererCPP>(
+            dynamic_cast<BertQuestionAnswererCPP*>(
+                bert_qa_status.value().release()))};
+  } else {
+    return nullptr;
+  }
+}
+
+QaAnswers* BertQuestionAnswererAnswer(
+    const BertQuestionAnswerer* question_answerer, const char* context,
+    const char* question) {
+  std::vector<QaAnswerCPP> answers = question_answerer->impl->Answer(
+      absl::string_view(context).data(), absl::string_view(question).data());
+  size_t size = answers.size();
+  auto* qa_answers = new QaAnswer[size];
+
+  for (size_t i = 0; i < size; ++i) {
+    qa_answers[i].start = answers[i].pos.start;
+    qa_answers[i].end = answers[i].pos.end;
+    qa_answers[i].logit = answers[i].pos.logit;
+    qa_answers[i].text = strdup(answers[i].text.c_str());
+  }
+
+  auto* c_answers = new QaAnswers;
+  c_answers->size = size;
+  c_answers->answers = qa_answers;
+  return c_answers;
+}
+
+void BertQuestionAnswererDelete(BertQuestionAnswerer* bert_question_answerer) {
+  delete bert_question_answerer;
+}
+
+void BertQuestionAnswererQaAnswersDelete(QaAnswers* qa_answers) {
+  delete[] qa_answers->answers;
+  delete qa_answers;
+}
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
diff --git a/tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.h b/tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.h
new file mode 100644
index 0000000..7fd3694
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.h
@@ -0,0 +1,78 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_BERT_QA_C_API_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_BERT_QA_C_API_H_
+
+// --------------------------------------------------------------------------
+/// C API for BertQuestionAnswerer.
+///
+/// The API leans towards simplicity and uniformity instead of convenience, as
+/// most usage will be by language-specific wrappers. It provides largely the
+/// same set of functionality as that of the C++ TensorFlow Lite
+/// `BertQuestionAnswerer` API, but is useful for shared libraries where having
+/// a stable ABI boundary is important.
+///
+/// Usage:
+/// <pre><code>
+/// // Create the model and interpreter options.
+/// BertQuestionAnswerer* qa_answerer =
+///   BertQuestionAnswererFromFile("/path/to/model.tflite");
+///
+/// // answer a question.
+/// QaAnswers* answers = Answer(qa_answerer, context, question);
+///
+/// // Dispose of the API and QaAnswers objects.
+/// BertQuestionAnswererDelete(qa_answerer);
+/// BertQuestionAnswererQaAnswersDelete(answers);
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+typedef struct BertQuestionAnswerer BertQuestionAnswerer;
+
+struct QaAnswer {
+  int start;
+  int end;
+  float logit;
+  char* text;
+};
+
+struct QaAnswers {
+  int size;
+  struct QaAnswer* answers;
+};
+
+// Creates BertQuestionAnswerer from model path, returns nullptr if the file
+// doesn't exist or is not a well formatted TFLite model path.
+extern BertQuestionAnswerer* BertQuestionAnswererFromFile(
+    const char* model_path);
+
+// Invokes the encapsulated TFLite model and answers a question based on
+// context.
+extern struct QaAnswers* BertQuestionAnswererAnswer(
+    const BertQuestionAnswerer* question_answerer, const char* context,
+    const char* question);
+
+extern void BertQuestionAnswererDelete(
+    BertQuestionAnswerer* bert_question_answerer);
+
+extern void BertQuestionAnswererQaAnswersDelete(struct QaAnswers* qa_answers);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif  // __cplusplus
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_BERT_QA_C_API_H_
diff --git a/tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.cc b/tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.cc
new file mode 100644
index 0000000..aa7ffef
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.cc
@@ -0,0 +1,393 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.h"
+
+#include "absl/strings/str_join.h"
+#include "absl/strings/str_split.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace qa {
+
+constexpr char kIdsTensorName[] = "ids";
+constexpr char kMaskTensorName[] = "mask";
+constexpr char kSegmentIdsTensorName[] = "segment_ids";
+constexpr char kEndLogitsTensorName[] = "end_logits";
+constexpr char kStartLogitsTensorName[] = "start_logits";
+
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+using ::tflite::support::text::tokenizer::BertTokenizer;
+using ::tflite::support::text::tokenizer::CreateTokenizerFromProcessUnit;
+using ::tflite::support::text::tokenizer::SentencePieceTokenizer;
+using ::tflite::support::text::tokenizer::TokenizerResult;
+using ::tflite::task::core::FindTensorByName;
+using ::tflite::task::core::PopulateTensor;
+using ::tflite::task::core::PopulateVector;
+using ::tflite::task::core::ReverseSortIndices;
+
+namespace {
+constexpr int kTokenizerProcessUnitIndex = 0;
+}
+
+StatusOr<std::unique_ptr<QuestionAnswerer>>
+BertQuestionAnswerer::CreateFromFile(
+    const std::string& path_to_model_with_metadata) {
+  std::unique_ptr<BertQuestionAnswerer> api_to_init;
+  ASSIGN_OR_RETURN(
+      api_to_init,
+      core::TaskAPIFactory::CreateFromFile<BertQuestionAnswerer>(
+          path_to_model_with_metadata,
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+          kNumLiteThreads));
+  RETURN_IF_ERROR(api_to_init->InitializeFromMetadata());
+  return api_to_init;
+}
+
+StatusOr<std::unique_ptr<QuestionAnswerer>>
+BertQuestionAnswerer::CreateFromBuffer(
+    const char* model_with_metadata_buffer_data,
+    size_t model_with_metadata_buffer_size) {
+  std::unique_ptr<BertQuestionAnswerer> api_to_init;
+  ASSIGN_OR_RETURN(
+      api_to_init,
+      core::TaskAPIFactory::CreateFromBuffer<BertQuestionAnswerer>(
+          model_with_metadata_buffer_data, model_with_metadata_buffer_size,
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+          kNumLiteThreads));
+  RETURN_IF_ERROR(api_to_init->InitializeFromMetadata());
+  return api_to_init;
+}
+
+StatusOr<std::unique_ptr<QuestionAnswerer>> BertQuestionAnswerer::CreateFromFd(
+    int fd) {
+  std::unique_ptr<BertQuestionAnswerer> api_to_init;
+  ASSIGN_OR_RETURN(
+      api_to_init,
+      core::TaskAPIFactory::CreateFromFileDescriptor<BertQuestionAnswerer>(
+          fd, absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+          kNumLiteThreads));
+  RETURN_IF_ERROR(api_to_init->InitializeFromMetadata());
+  return api_to_init;
+}
+
+StatusOr<std::unique_ptr<QuestionAnswerer>>
+BertQuestionAnswerer::CreateBertQuestionAnswererFromFile(
+    const std::string& path_to_model, const std::string& path_to_vocab) {
+  std::unique_ptr<BertQuestionAnswerer> api_to_init;
+  ASSIGN_OR_RETURN(
+      api_to_init,
+      core::TaskAPIFactory::CreateFromFile<BertQuestionAnswerer>(
+          path_to_model,
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+          kNumLiteThreads));
+  api_to_init->InitializeBertTokenizer(path_to_vocab);
+  return api_to_init;
+}
+
+StatusOr<std::unique_ptr<QuestionAnswerer>>
+BertQuestionAnswerer::CreateBertQuestionAnswererFromBuffer(
+    const char* model_buffer_data, size_t model_buffer_size,
+    const char* vocab_buffer_data, size_t vocab_buffer_size) {
+  std::unique_ptr<BertQuestionAnswerer> api_to_init;
+  ASSIGN_OR_RETURN(
+      api_to_init,
+      core::TaskAPIFactory::CreateFromBuffer<BertQuestionAnswerer>(
+          model_buffer_data, model_buffer_size,
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+          kNumLiteThreads));
+  api_to_init->InitializeBertTokenizerFromBinary(vocab_buffer_data,
+                                                 vocab_buffer_size);
+  return api_to_init;
+}
+
+StatusOr<std::unique_ptr<QuestionAnswerer>>
+BertQuestionAnswerer::CreateAlbertQuestionAnswererFromFile(
+    const std::string& path_to_model, const std::string& path_to_spmodel) {
+  std::unique_ptr<BertQuestionAnswerer> api_to_init;
+  ASSIGN_OR_RETURN(
+      api_to_init,
+      core::TaskAPIFactory::CreateFromFile<BertQuestionAnswerer>(
+          path_to_model,
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+          kNumLiteThreads));
+  api_to_init->InitializeSentencepieceTokenizer(path_to_spmodel);
+  return api_to_init;
+}
+
+StatusOr<std::unique_ptr<QuestionAnswerer>>
+BertQuestionAnswerer::CreateAlbertQuestionAnswererFromBuffer(
+    const char* model_buffer_data, size_t model_buffer_size,
+    const char* spmodel_buffer_data, size_t spmodel_buffer_size) {
+  std::unique_ptr<BertQuestionAnswerer> api_to_init;
+  ASSIGN_OR_RETURN(
+      api_to_init,
+      core::TaskAPIFactory::CreateFromBuffer<BertQuestionAnswerer>(
+          model_buffer_data, model_buffer_size,
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>(),
+          kNumLiteThreads));
+  api_to_init->InitializeSentencepieceTokenizerFromBinary(spmodel_buffer_data,
+                                                          spmodel_buffer_size);
+  return api_to_init;
+}
+
+std::vector<QaAnswer> BertQuestionAnswerer::Answer(
+    const std::string& context, const std::string& question) {
+  // The BertQuestionAnswererer implementation for Preprocess() and
+  // Postprocess() never returns errors: just call value().
+  return Infer(context, question).value();
+}
+
+absl::Status BertQuestionAnswerer::Preprocess(
+    const std::vector<TfLiteTensor*>& input_tensors, const std::string& context,
+    const std::string& query) {
+  auto* input_tensor_metadatas =
+      GetMetadataExtractor()->GetInputTensorMetadata();
+  TfLiteTensor* ids_tensor =
+      input_tensor_metadatas
+          ? FindTensorByName(input_tensors, input_tensor_metadatas,
+                             kIdsTensorName)
+          : input_tensors[0];
+  TfLiteTensor* mask_tensor =
+      input_tensor_metadatas
+          ? FindTensorByName(input_tensors, input_tensor_metadatas,
+                             kMaskTensorName)
+          : input_tensors[1];
+  TfLiteTensor* segment_ids_tensor =
+      input_tensor_metadatas
+          ? FindTensorByName(input_tensors, input_tensor_metadatas,
+                             kSegmentIdsTensorName)
+          : input_tensors[2];
+
+  token_to_orig_map_.clear();
+
+  // The orig_tokens is used for recovering the answer string from the index,
+  // while the processed_tokens is lower-cased and used to generate input of
+  // the model.
+  orig_tokens_ = absl::StrSplit(context, absl::ByChar(' '), absl::SkipEmpty());
+  std::vector<std::string> processed_tokens(orig_tokens_);
+
+  std::string processed_query = query;
+  if (kUseLowerCase) {
+    for (auto& token : processed_tokens) {
+      absl::AsciiStrToLower(&token);
+    }
+    absl::AsciiStrToLower(&processed_query);
+  }
+
+  TokenizerResult query_tokenize_results;
+  query_tokenize_results = tokenizer_->Tokenize(processed_query);
+
+  std::vector<std::string> query_tokens = query_tokenize_results.subwords;
+  if (query_tokens.size() > kMaxQueryLen) {
+    query_tokens.resize(kMaxQueryLen);
+  }
+
+  // Example:
+  // context:             tokenize     me  please
+  // all_doc_tokens:      token ##ize  me  plea ##se
+  // token_to_orig_index: [0,   0,     1,  2,   2]
+
+  std::vector<std::string> all_doc_tokens;
+  std::vector<int> token_to_orig_index;
+  for (size_t i = 0; i < processed_tokens.size(); i++) {
+    const std::string& token = processed_tokens[i];
+    std::vector<std::string> sub_tokens = tokenizer_->Tokenize(token).subwords;
+    for (const std::string& sub_token : sub_tokens) {
+      token_to_orig_index.emplace_back(i);
+      all_doc_tokens.emplace_back(sub_token);
+    }
+  }
+
+  // -3 accounts for [CLS], [SEP] and [SEP].
+  int max_context_len = kMaxSeqLen - query_tokens.size() - 3;
+  if (all_doc_tokens.size() > max_context_len) {
+    all_doc_tokens.resize(max_context_len);
+  }
+
+  std::vector<std::string> tokens;
+  tokens.reserve(3 + query_tokens.size() + all_doc_tokens.size());
+  std::vector<int> segment_ids;
+  segment_ids.reserve(kMaxSeqLen);
+
+  // Start of generating the features.
+  tokens.emplace_back("[CLS]");
+  segment_ids.emplace_back(0);
+
+  // For query input.
+  for (const auto& query_token : query_tokens) {
+    tokens.emplace_back(query_token);
+    segment_ids.emplace_back(0);
+  }
+
+  // For Separation.
+  tokens.emplace_back("[SEP]");
+  segment_ids.emplace_back(0);
+
+  // For Text Input.
+  for (int i = 0; i < all_doc_tokens.size(); i++) {
+    auto& doc_token = all_doc_tokens[i];
+    tokens.emplace_back(doc_token);
+    segment_ids.emplace_back(1);
+    token_to_orig_map_[tokens.size()] = token_to_orig_index[i];
+  }
+
+  // For ending mark.
+  tokens.emplace_back("[SEP]");
+  segment_ids.emplace_back(1);
+
+  std::vector<int> input_ids(tokens.size());
+  input_ids.reserve(kMaxSeqLen);
+  // Convert tokens back into ids
+  for (int i = 0; i < tokens.size(); i++) {
+    auto& token = tokens[i];
+    tokenizer_->LookupId(token, &input_ids[i]);
+  }
+
+  std::vector<int> input_mask;
+  input_mask.reserve(kMaxSeqLen);
+  input_mask.insert(input_mask.end(), tokens.size(), 1);
+
+  int zeros_to_pad = kMaxSeqLen - input_ids.size();
+  input_ids.insert(input_ids.end(), zeros_to_pad, 0);
+  input_mask.insert(input_mask.end(), zeros_to_pad, 0);
+  segment_ids.insert(segment_ids.end(), zeros_to_pad, 0);
+
+  // input_ids INT32[1, 384]
+  PopulateTensor(input_ids, ids_tensor);
+  // input_mask INT32[1, 384]
+  PopulateTensor(input_mask, mask_tensor);
+  // segment_ids INT32[1, 384]
+  PopulateTensor(segment_ids, segment_ids_tensor);
+
+  return absl::OkStatus();
+}
+
+StatusOr<std::vector<QaAnswer>> BertQuestionAnswerer::Postprocess(
+    const std::vector<const TfLiteTensor*>& output_tensors,
+    const std::string& /*lowercased_context*/,
+    const std::string& /*lowercased_query*/) {
+  auto* output_tensor_metadatas =
+      GetMetadataExtractor()->GetOutputTensorMetadata();
+
+  const TfLiteTensor* end_logits_tensor =
+      output_tensor_metadatas
+          ? FindTensorByName(output_tensors, output_tensor_metadatas,
+                             kEndLogitsTensorName)
+          : output_tensors[0];
+  const TfLiteTensor* start_logits_tensor =
+      output_tensor_metadatas
+          ? FindTensorByName(output_tensors, output_tensor_metadatas,
+                             kStartLogitsTensorName)
+          : output_tensors[1];
+
+  std::vector<float> end_logits;
+  std::vector<float> start_logits;
+
+  // end_logits FLOAT[1, 384]
+  PopulateVector(end_logits_tensor, &end_logits);
+  // start_logits FLOAT[1, 384]
+  PopulateVector(start_logits_tensor, &start_logits);
+
+  auto start_indices = ReverseSortIndices(start_logits);
+  auto end_indices = ReverseSortIndices(end_logits);
+
+  std::vector<QaAnswer::Pos> orig_results;
+  for (int start_index = 0; start_index < kPredictAnsNum; start_index++) {
+    for (int end_index = 0; end_index < kPredictAnsNum; end_index++) {
+      int start = start_indices[start_index];
+      int end = end_indices[end_index];
+
+      if (!token_to_orig_map_.contains(start + kOutputOffset) ||
+          !token_to_orig_map_.contains(end + kOutputOffset) || end < start ||
+          (end - start + 1) > kMaxAnsLen) {
+        continue;
+      }
+      orig_results.emplace_back(
+          QaAnswer::Pos(start, end, start_logits[start] + end_logits[end]));
+    }
+  }
+
+  std::sort(orig_results.begin(), orig_results.end());
+
+  std::vector<QaAnswer> answers;
+  for (int i = 0; i < orig_results.size() && i < kPredictAnsNum; i++) {
+    auto orig_pos = orig_results[i];
+    answers.emplace_back(
+        orig_pos.start > 0 ? ConvertIndexToString(orig_pos.start, orig_pos.end)
+                           : "",
+        orig_pos);
+  }
+
+  return answers;
+}
+
+std::string BertQuestionAnswerer::ConvertIndexToString(int start, int end) {
+  int start_index = token_to_orig_map_[start + kOutputOffset];
+  int end_index = token_to_orig_map_[end + kOutputOffset];
+
+  return absl::StrJoin(orig_tokens_.begin() + start_index,
+                       orig_tokens_.begin() + end_index + 1, " ");
+}
+
+absl::Status BertQuestionAnswerer::InitializeFromMetadata() {
+  const ProcessUnit* tokenizer_process_unit =
+      GetMetadataExtractor()->GetInputProcessUnit(kTokenizerProcessUnitIndex);
+  if (tokenizer_process_unit == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "No input process unit found from metadata.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+  ASSIGN_OR_RETURN(tokenizer_,
+                   CreateTokenizerFromProcessUnit(tokenizer_process_unit,
+                                                  GetMetadataExtractor()));
+  return absl::OkStatus();
+}
+
+void BertQuestionAnswerer::InitializeBertTokenizer(
+    const std::string& path_to_vocab) {
+  tokenizer_ = absl::make_unique<BertTokenizer>(path_to_vocab);
+}
+
+void BertQuestionAnswerer::InitializeBertTokenizerFromBinary(
+    const char* vocab_buffer_data, size_t vocab_buffer_size) {
+  tokenizer_ =
+      absl::make_unique<BertTokenizer>(vocab_buffer_data, vocab_buffer_size);
+}
+
+void BertQuestionAnswerer::InitializeSentencepieceTokenizer(
+    const std::string& path_to_spmodel) {
+  tokenizer_ = absl::make_unique<SentencePieceTokenizer>(path_to_spmodel);
+}
+
+void BertQuestionAnswerer::InitializeSentencepieceTokenizerFromBinary(
+    const char* spmodel_buffer_data, size_t spmodel_buffer_size) {
+  tokenizer_ = absl::make_unique<SentencePieceTokenizer>(spmodel_buffer_data,
+                                                         spmodel_buffer_size);
+}
+
+}  // namespace qa
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.h b/tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.h
new file mode 100644
index 0000000..4c65dc0
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.h
@@ -0,0 +1,170 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_BERT_QUESTION_ANSWERER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_BERT_QUESTION_ANSWERER_H_
+
+#include "absl/container/flat_hash_map.h"
+#include "absl/status/status.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/base_task_api.h"
+#include "tensorflow_lite_support/cc/task/core/task_api_factory.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+#include "tensorflow_lite_support/cc/task/text/qa/question_answerer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace qa {
+
+// BertQA task API, performs tokenization for models (BERT, Albert, etc.) in
+// preprocess and returns most possible answers.
+//
+// In particular, the branch of BERT models use WordPiece tokenizer, and the
+// branch of Albert models use SentencePiece tokenizer, respectively.
+//
+// Factory methods:
+//   CreateFromFile(path_to_model_with_metadata)
+//   CreateFromBuffer(model_with_metadata_buffer_data,
+//                                model_with_metadata_buffer_size)
+//   CreateFromFd(file_descriptor_to_model_with_metadata)
+//     Generic API to create the QuestionAnswerer for bert models with metadata
+//     populated. The API expects a Bert based TFLite model with metadata
+//     containing the following information:
+//       - input_process_units for Wordpiece/Sentencepiece Tokenizer. Wordpiece
+//         Tokenizer can be used for a MobileBert[0] model, Sentencepiece
+//         Tokenizer Tokenizer can be used for an Albert[1] model
+//       - 3 input tensors with names "ids", "mask" and "segment_ids"
+//       - 2 output tensors with names "end_logits" and "start_logits"
+//      [0]: https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1
+//      [1]: https://tfhub.dev/tensorflow/lite-model/albert_lite_base/squadv1/1
+//
+//   CreateBertQuestionAnswererFromFile(path_to_model, path_to_vocab)
+//     Creates a BertQuestionAnswerer from TFLite model file and vocab file for
+//     WordPiece tokenizer. Used in C++ environment.
+//     One suitable model is:
+//       https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1
+//
+//   CreateBertQuestionAnswererFromBuffer(model_buffer_data, model_buffer_size,
+//                                        vocab_buffer_data, vocab_buffer_size)
+//     Creates a BertQuestionAnswerer from TFLite model buffer and vocab file
+//     buffer for WordPiece tokenizer. Used in Jave (JNI) environment.
+//
+//   CreateAlbertQuestionAnswererFromFile(path_to_model, path_to_spmodel)
+//     Creates an AlbertQuestionAnswerer from TFLite model file and
+//     SentencePiece model file. Used in C++ environment.
+//     One suitable model is:
+//       https://tfhub.dev/tensorflow/lite-model/albert_lite_base/squadv1/1
+//
+//   CreateAlbertQuestionAnswererFromBuffer(model_buffer_data,
+//                                          model_buffer_size,
+//                                          spmodel_buffer_data,
+//                                          spmodel_buffer_size)
+//     Creates an AlbertQuestionAnswerer from TFLite model file buffer and
+//     SentencePiece model file buffer. Used in Jave (JNI) environment.
+//
+
+class BertQuestionAnswerer : public QuestionAnswerer {
+ public:
+  // TODO(b/150904655): add support to parameterize.
+  static constexpr int kMaxQueryLen = 64;
+  static constexpr int kMaxSeqLen = 384;
+  static constexpr int kPredictAnsNum = 5;
+  static constexpr int kMaxAnsLen = 32;
+  // TODO(b/151954803): clarify the offset usage
+  static constexpr int kOutputOffset = 1;
+  static constexpr int kNumLiteThreads = 4;
+  static constexpr bool kUseLowerCase = true;
+
+  static tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>>
+  CreateFromFile(const std::string& path_to_model_with_metadata);
+
+  static tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>>
+  CreateFromBuffer(const char* model_with_metadata_buffer_data,
+                   size_t model_with_metadata_buffer_size);
+
+  static tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>>
+  CreateFromFd(int fd);
+
+  static tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>>
+  CreateBertQuestionAnswererFromFile(const std::string& path_to_model,
+                                     const std::string& path_to_vocab);
+
+  static tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>>
+  CreateBertQuestionAnswererFromBuffer(const char* model_buffer_data,
+                                       size_t model_buffer_size,
+                                       const char* vocab_buffer_data,
+                                       size_t vocab_buffer_size);
+
+  static tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>>
+  CreateAlbertQuestionAnswererFromFile(const std::string& path_to_model,
+                                       const std::string& path_to_spmodel);
+
+  static tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>>
+  CreateAlbertQuestionAnswererFromBuffer(const char* model_buffer_data,
+                                         size_t model_buffer_size,
+                                         const char* spmodel_buffer_data,
+                                         size_t spmodel_buffer_size);
+
+  explicit BertQuestionAnswerer(std::unique_ptr<core::TfLiteEngine> engine)
+      : QuestionAnswerer(std::move(engine)) {}
+
+  // Answers question based on the context. Could be empty if no answer was
+  // found from the given context.
+  std::vector<QaAnswer> Answer(const std::string& context,
+                               const std::string& question) override;
+
+ private:
+  absl::Status Preprocess(const std::vector<TfLiteTensor*>& input_tensors,
+                          const std::string& lowercased_context,
+                          const std::string& lowercased_query) override;
+
+  tflite::support::StatusOr<std::vector<QaAnswer>> Postprocess(
+      const std::vector<const TfLiteTensor*>& output_tensors,
+      const std::string& lowercased_context,
+      const std::string& lowercased_query) override;
+
+  // Initialize API with a BertTokenizer from the vocabulary file.
+  void InitializeBertTokenizer(const std::string& path_to_vocab);
+  // Initialize API with a BertTokenizer from the vocabulary buffer.
+  void InitializeBertTokenizerFromBinary(const char* vocab_buffer_data,
+                                         size_t vocab_buffer_size);
+
+  // Initialize API with a SentencepieceTokenizer from the model file.
+  void InitializeSentencepieceTokenizer(const std::string& path_to_spmodel);
+  // Initialize API with a SentencepieceTokenizer from the model buffer.
+  void InitializeSentencepieceTokenizerFromBinary(
+      const char* spmodel_buffer_data, size_t spmodel_buffer_size);
+
+  // Initialize the API with the tokenizer set in the metadata.
+  absl::Status InitializeFromMetadata();
+
+  std::string ConvertIndexToString(int start, int end);
+
+  std::unique_ptr<tflite::support::text::tokenizer::Tokenizer> tokenizer_;
+  // Maps index of input token to index of untokenized word from original input.
+  absl::flat_hash_map<size_t, size_t> token_to_orig_map_;
+  // Original tokens of context.
+  std::vector<std::string> orig_tokens_;
+};
+
+}  // namespace qa
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_BERT_QUESTION_ANSWERER_H_
diff --git a/tensorflow_lite_support/cc/task/text/qa/question_answerer.h b/tensorflow_lite_support/cc/task/text/qa/question_answerer.h
new file mode 100644
index 0000000..f46a40c
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/text/qa/question_answerer.h
@@ -0,0 +1,65 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_QUESTION_ANSWERER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_QUESTION_ANSWERER_H_
+
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "tensorflow_lite_support/cc/task/core/base_task_api.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace qa {
+
+// Struct for the Answer to QuestionAnswerer.
+struct QaAnswer {
+  // struct to represent the logit and offset of the answer related to context.
+  struct Pos {
+    Pos(int arg_start, int arg_end, float arg_logit)
+        : start(arg_start), end(arg_end), logit(arg_logit) {}
+    int start, end;
+    float logit;
+    bool operator<(const Pos& rhs) const { return rhs.logit < logit; }
+  };
+
+  QaAnswer(std::string arg_text, Pos arg_pos)
+      : text(std::move(arg_text)), pos(arg_pos) {}
+  std::string text;
+  Pos pos;
+};
+
+// Interface for an Question-Answer API.
+class QuestionAnswerer
+    : public core::BaseTaskApi<std::vector<QaAnswer>, const std::string&,
+                               const std::string&> {
+ public:
+  explicit QuestionAnswerer(std::unique_ptr<core::TfLiteEngine> engine)
+      : BaseTaskApi(std::move(engine)) {}
+
+  virtual std::vector<QaAnswer> Answer(const std::string& context,
+                                       const std::string& question) = 0;
+};
+
+}  // namespace qa
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_TEXT_QA_QUESTION_ANSWERER_H_
diff --git a/tensorflow_lite_support/cc/task/vision/BUILD b/tensorflow_lite_support/cc/task/vision/BUILD
new file mode 100644
index 0000000..d426486
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/BUILD
@@ -0,0 +1,108 @@
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "object_detector",
+    srcs = ["object_detector.cc"],
+    hdrs = ["object_detector.h"],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:external_file_handler",
+        "//tensorflow_lite_support/cc/task/core:task_api_factory",
+        "//tensorflow_lite_support/cc/task/core:task_utils",
+        "//tensorflow_lite_support/cc/task/core:tflite_engine",
+        "//tensorflow_lite_support/cc/task/vision/core:base_vision_task_api",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/core:label_map_item",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:class_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:detections_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:object_detector_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_utils",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/container:flat_hash_set",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api",
+    ],
+)
+
+cc_library(
+    name = "image_classifier",
+    srcs = ["image_classifier.cc"],
+    hdrs = ["image_classifier.h"],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:external_file_handler",
+        "//tensorflow_lite_support/cc/task/core:task_api_factory",
+        "//tensorflow_lite_support/cc/task/core:task_utils",
+        "//tensorflow_lite_support/cc/task/core:tflite_engine",
+        "//tensorflow_lite_support/cc/task/vision/core:base_vision_task_api",
+        "//tensorflow_lite_support/cc/task/vision/core:classification_head",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/core:label_map_item",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:class_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:classifications_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_classifier_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_utils",
+        "//tensorflow_lite_support/cc/task/vision/utils:score_calibration",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/algorithm:container",
+        "@com_google_absl//absl/container:flat_hash_set",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api",
+    ],
+)
+
+cc_library(
+    name = "image_segmenter",
+    srcs = ["image_segmenter.cc"],
+    hdrs = ["image_segmenter.h"],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:external_file_handler",
+        "//tensorflow_lite_support/cc/task/core:task_api_factory",
+        "//tensorflow_lite_support/cc/task/core:task_utils",
+        "//tensorflow_lite_support/cc/task/core:tflite_engine",
+        "//tensorflow_lite_support/cc/task/vision/core:base_vision_task_api",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/core:label_map_item",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_segmenter_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:segmentations_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_utils",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/core/api",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/task/vision/core/BUILD b/tensorflow_lite_support/cc/task/vision/core/BUILD
new file mode 100644
index 0000000..1df86cb
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/BUILD
@@ -0,0 +1,81 @@
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files(srcs = ["base_vision_task_api.h"])
+
+cc_library(
+    name = "base_vision_task_api",
+    hdrs = [
+        "base_vision_task_api.h",
+    ],
+    deps = [
+        ":frame_buffer",
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/task/core:base_task_api",
+        "//tensorflow_lite_support/cc/task/core:task_utils",
+        "//tensorflow_lite_support/cc/task/core:tflite_engine",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_utils",
+        "//tensorflow_lite_support/cc/task/vision/utils:image_tensor_specs",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/time",
+        "@org_tensorflow//tensorflow/lite/c:common",
+    ],
+)
+
+cc_library(
+    name = "frame_buffer",
+    srcs = ["frame_buffer.cc"],
+    hdrs = ["frame_buffer.h"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/time",
+        "@com_google_absl//absl/types:any",
+        "@com_google_absl//absl/types:optional",
+    ],
+)
+
+cc_library(
+    name = "label_map_item",
+    srcs = ["label_map_item.cc"],
+    hdrs = ["label_map_item.h"],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "@com_google_absl//absl/container:flat_hash_map",
+        "@com_google_absl//absl/container:flat_hash_set",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
+
+cc_library(
+    name = "classification_head",
+    srcs = ["classification_head.cc"],
+    hdrs = ["classification_head.h"],
+    deps = [
+        ":label_map_item",
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision/utils:score_calibration",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/task/vision/core/base_vision_task_api.h b/tensorflow_lite_support/cc/task/vision/core/base_vision_task_api.h
new file mode 100644
index 0000000..feb6b4a
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/base_vision_task_api.h
@@ -0,0 +1,270 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_BASE_VISION_TASK_API_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_BASE_VISION_TASK_API_H_
+
+#include <array>
+#include <memory>
+#include <utility>
+#include <vector>
+
+#include "absl/memory/memory.h"
+#include "absl/status/status.h"
+#include "absl/time/clock.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/core/base_task_api.h"
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Base class providing common logic for vision models.
+template <class OutputType>
+class BaseVisionTaskApi
+    : public tflite::task::core::BaseTaskApi<OutputType, const FrameBuffer&,
+                                             const BoundingBox&> {
+ public:
+  explicit BaseVisionTaskApi(std::unique_ptr<core::TfLiteEngine> engine)
+      : tflite::task::core::BaseTaskApi<OutputType, const FrameBuffer&,
+                                        const BoundingBox&>(std::move(engine)) {
+  }
+  // BaseVisionTaskApi is neither copyable nor movable.
+  BaseVisionTaskApi(const BaseVisionTaskApi&) = delete;
+  BaseVisionTaskApi& operator=(const BaseVisionTaskApi&) = delete;
+
+  // Number of bytes required for 8-bit per pixel RGB color space.
+  static constexpr int kRgbPixelBytes = 3;
+
+  // Sets the ProcessEngine used for image pre-processing. Must be called before
+  // any inference is performed. Can be called between inferences to override
+  // the current process engine.
+  void SetProcessEngine(const FrameBufferUtils::ProcessEngine& process_engine) {
+    frame_buffer_utils_ = FrameBufferUtils::Create(process_engine);
+  }
+
+ protected:
+  using tflite::task::core::BaseTaskApi<OutputType, const FrameBuffer&,
+                                        const BoundingBox&>::engine_;
+
+  // Checks input tensor and metadata (if any) are valid, or return an error
+  // otherwise. This must be called once at initialization time, before running
+  // inference, as it is a prerequisite for `Preprocess`.
+  // Note: the underlying interpreter and metadata extractor are assumed to be
+  // already successfully initialized before calling this method.
+  virtual absl::Status CheckAndSetInputs() {
+    ASSIGN_OR_RETURN(
+        ImageTensorSpecs input_specs,
+        BuildInputImageTensorSpecs(*engine_->interpreter(),
+                                   *engine_->metadata_extractor()));
+
+    if (input_specs.color_space != tflite::ColorSpaceType_RGB) {
+      return tflite::support::CreateStatusWithPayload(
+          absl::StatusCode::kUnimplemented,
+          "BaseVisionTaskApi only supports RGB color space for now.");
+    }
+
+    input_specs_ = absl::make_unique<ImageTensorSpecs>(input_specs);
+
+    return absl::OkStatus();
+  }
+
+  // Performs image preprocessing on the input frame buffer over the region of
+  // interest so that it fits model requirements (e.g. upright 224x224 RGB) and
+  // populate the corresponding input tensor. This is performed by (in this
+  // order):
+  // - cropping the frame buffer to the region of interest (which, in most
+  //   cases, just covers the entire input image),
+  // - resizing it (with bilinear interpolation, aspect-ratio *not* preserved)
+  //   to the dimensions of the model input tensor,
+  // - converting it to the colorspace of the input tensor (i.e. RGB, which is
+  //   the only supported colorspace for now),
+  // - rotating it according to its `Orientation` so that inference is performed
+  //   on an "upright" image.
+  //
+  // IMPORTANT: as a consequence of cropping occurring first, the provided
+  // region of interest is expressed in the unrotated frame of reference
+  // coordinates system, i.e. in `[0, frame_buffer.width) x [0,
+  // frame_buffer.height)`, which are the dimensions of the underlying
+  // `frame_buffer` data before any `Orientation` flag gets applied. Also, the
+  // region of interest is not clamped, so this method will return a non-ok
+  // status if the region is out of these bounds.
+  absl::Status Preprocess(const std::vector<TfLiteTensor*>& input_tensors,
+                          const FrameBuffer& frame_buffer,
+                          const BoundingBox& roi) override {
+    if (input_specs_ == nullptr) {
+      return tflite::support::CreateStatusWithPayload(
+          absl::StatusCode::kInternal,
+          "Uninitialized input tensor specs: CheckAndSetInputs must be called "
+          "at initialization time.");
+    }
+
+    if (frame_buffer_utils_ == nullptr) {
+      return tflite::support::CreateStatusWithPayload(
+          absl::StatusCode::kInternal,
+          "Uninitialized frame buffer utils: SetProcessEngine must be called "
+          "at initialization time.");
+    }
+
+    if (input_tensors.size() != 1) {
+      return tflite::support::CreateStatusWithPayload(
+          absl::StatusCode::kInternal, "A single input tensor is expected.");
+    }
+
+    // Input data to be normalized (if needed) and used for inference. In most
+    // cases, this is the result of image preprocessing. In case no image
+    // preprocessing is needed (see below), this points to the input frame
+    // buffer raw data.
+    const uint8* input_data;
+    size_t input_data_byte_size;
+
+    // Optional buffers in case image preprocessing is needed.
+    std::unique_ptr<FrameBuffer> preprocessed_frame_buffer;
+    std::vector<uint8> preprocessed_data;
+
+    if (IsImagePreprocessingNeeded(frame_buffer, roi)) {
+      // Preprocess input image to fit model requirements.
+      // For now RGB is the only color space supported, which is ensured by
+      // `CheckAndSetInputs`.
+      FrameBuffer::Dimension to_buffer_dimension = {input_specs_->image_width,
+                                                    input_specs_->image_height};
+      input_data_byte_size =
+          GetBufferByteSize(to_buffer_dimension, FrameBuffer::Format::kRGB);
+      preprocessed_data.resize(input_data_byte_size / sizeof(uint8), 0);
+      input_data = preprocessed_data.data();
+
+      FrameBuffer::Plane preprocessed_plane = {
+          /*buffer=*/preprocessed_data.data(),
+          /*stride=*/{input_specs_->image_width * kRgbPixelBytes,
+                      kRgbPixelBytes}};
+      preprocessed_frame_buffer = FrameBuffer::Create(
+          {preprocessed_plane}, to_buffer_dimension, FrameBuffer::Format::kRGB,
+          FrameBuffer::Orientation::kTopLeft);
+
+      RETURN_IF_ERROR(frame_buffer_utils_->Preprocess(
+          frame_buffer, roi, preprocessed_frame_buffer.get()));
+    } else {
+      // Input frame buffer already targets model requirements: skip image
+      // preprocessing. For RGB, the data is always stored in a single plane.
+      input_data = frame_buffer.plane(0).buffer;
+      input_data_byte_size = frame_buffer.plane(0).stride.row_stride_bytes *
+                             frame_buffer.dimension().height;
+    }
+
+    // Then normalize pixel data (if needed) and populate the input tensor.
+    switch (input_specs_->tensor_type) {
+      case kTfLiteUInt8:
+        if (input_tensors[0]->bytes != input_data_byte_size) {
+          return tflite::support::CreateStatusWithPayload(
+              absl::StatusCode::kInternal,
+              "Size mismatch or unsupported padding bytes between pixel data "
+              "and input tensor.");
+        }
+        // No normalization required: directly populate data.
+        tflite::task::core::PopulateTensor(
+            input_data, input_data_byte_size / sizeof(uint8), input_tensors[0]);
+        break;
+      case kTfLiteFloat32: {
+        if (input_tensors[0]->bytes / sizeof(float) !=
+            input_data_byte_size / sizeof(uint8)) {
+          return tflite::support::CreateStatusWithPayload(
+              absl::StatusCode::kInternal,
+              "Size mismatch or unsupported padding bytes between pixel data "
+              "and input tensor.");
+        }
+        // Normalize and populate.
+        float* normalized_input_data =
+            tflite::task::core::AssertAndReturnTypedTensor<float>(
+                input_tensors[0]);
+        const tflite::task::vision::NormalizationOptions&
+            normalization_options = input_specs_->normalization_options.value();
+        if (normalization_options.num_values == 1) {
+          float mean_value = normalization_options.mean_values[0];
+          float inv_std_value = (1.0f / normalization_options.std_values[0]);
+          for (int i = 0; i < input_data_byte_size / sizeof(uint8);
+               i++, input_data++, normalized_input_data++) {
+            *normalized_input_data =
+                inv_std_value * (static_cast<float>(*input_data) - mean_value);
+          }
+        } else {
+          std::array<float, 3> inv_std_values = {
+              1.0f / normalization_options.std_values[0],
+              1.0f / normalization_options.std_values[1],
+              1.0f / normalization_options.std_values[2]};
+          for (int i = 0; i < input_data_byte_size / sizeof(uint8);
+               i++, input_data++, normalized_input_data++) {
+            *normalized_input_data = inv_std_values[i % 3] *
+                                     (static_cast<float>(*input_data) -
+                                      normalization_options.mean_values[i % 3]);
+          }
+        }
+        break;
+      }
+      case kTfLiteInt8:
+        return tflite::support::CreateStatusWithPayload(
+            absl::StatusCode::kUnimplemented,
+            "kTfLiteInt8 input type is not implemented yet.");
+      default:
+        return tflite::support::CreateStatusWithPayload(
+            absl::StatusCode::kInternal, "Unexpected input tensor type.");
+    }
+
+    return absl::OkStatus();
+  }
+
+  // Utils for input image preprocessing (resizing, colorspace conversion, etc).
+  std::unique_ptr<FrameBufferUtils> frame_buffer_utils_;
+
+  // Parameters related to the input tensor which represents an image.
+  std::unique_ptr<ImageTensorSpecs> input_specs_;
+
+ private:
+  // Returns false if image preprocessing could be skipped, true otherwise.
+  bool IsImagePreprocessingNeeded(const FrameBuffer& frame_buffer,
+                                  const BoundingBox& roi) {
+    // Is crop required?
+    if (roi.origin_x() != 0 || roi.origin_y() != 0 ||
+        roi.width() != frame_buffer.dimension().width ||
+        roi.height() != frame_buffer.dimension().height) {
+      return true;
+    }
+
+    // Are image transformations required?
+    if (frame_buffer.orientation() != FrameBuffer::Orientation::kTopLeft ||
+        frame_buffer.format() != FrameBuffer::Format::kRGB ||
+        frame_buffer.dimension().width != input_specs_->image_width ||
+        frame_buffer.dimension().height != input_specs_->image_height) {
+      return true;
+    }
+
+    return false;
+  }
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_BASE_VISION_TASK_API_H_
diff --git a/tensorflow_lite_support/cc/task/vision/core/classification_head.cc b/tensorflow_lite_support/cc/task/vision/core/classification_head.cc
new file mode 100644
index 0000000..962cb34
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/classification_head.cc
@@ -0,0 +1,114 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/cc/task/vision/core/classification_head.h"
+
+#include "absl/status/status.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+using ::absl::StatusCode;
+using ::tflite::metadata::ModelMetadataExtractor;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+
+StatusOr<ClassificationHead> BuildClassificationHead(
+    const tflite::metadata::ModelMetadataExtractor& metadata_extractor,
+    const tflite::TensorMetadata& output_tensor_metadata,
+    absl::string_view display_names_locale) {
+  ClassificationHead head;
+  if (output_tensor_metadata.name() != nullptr) {
+    head.name = output_tensor_metadata.name()->str();
+  }
+
+  // Build label map, if present.
+  const std::string labels_filename =
+      ModelMetadataExtractor::FindFirstAssociatedFileName(
+          output_tensor_metadata,
+          tflite::AssociatedFileType_TENSOR_AXIS_LABELS);
+  if (!labels_filename.empty()) {
+    ASSIGN_OR_RETURN(absl::string_view labels_file,
+                     metadata_extractor.GetAssociatedFile(labels_filename));
+    const std::string display_names_filename =
+        ModelMetadataExtractor::FindFirstAssociatedFileName(
+            output_tensor_metadata,
+            tflite::AssociatedFileType_TENSOR_AXIS_LABELS,
+            display_names_locale);
+    absl::string_view display_names_file;
+    if (!display_names_filename.empty()) {
+      ASSIGN_OR_RETURN(display_names_file, metadata_extractor.GetAssociatedFile(
+                                               display_names_filename));
+    }
+    ASSIGN_OR_RETURN(head.label_map_items,
+                     BuildLabelMapFromFiles(labels_file, display_names_file));
+  }
+
+  // Set score threshold, if present.
+  ASSIGN_OR_RETURN(const tflite::ProcessUnit* score_thresholding_process_unit,
+                   ModelMetadataExtractor::FindFirstProcessUnit(
+                       output_tensor_metadata,
+                       tflite::ProcessUnitOptions_ScoreThresholdingOptions));
+  if (score_thresholding_process_unit != nullptr) {
+    head.score_threshold =
+        score_thresholding_process_unit->options_as_ScoreThresholdingOptions()
+            ->global_score_threshold();
+  }
+
+  // Build score calibration parameters, if present.
+  ASSIGN_OR_RETURN(const tflite::ProcessUnit* score_calibration_process_unit,
+                   ModelMetadataExtractor::FindFirstProcessUnit(
+                       output_tensor_metadata,
+                       tflite::ProcessUnitOptions_ScoreCalibrationOptions));
+  if (score_calibration_process_unit != nullptr) {
+    if (labels_filename.empty()) {
+      return CreateStatusWithPayload(
+          StatusCode::kNotFound,
+          "Using ScoreCalibrationOptions requires a label map to be provided "
+          "as TENSOR_AXIS_LABELS associated file.",
+          TfLiteSupportStatus::kMetadataAssociatedFileNotFoundError);
+    }
+    const std::string score_calibration_filename =
+        ModelMetadataExtractor::FindFirstAssociatedFileName(
+            output_tensor_metadata,
+            tflite::AssociatedFileType_TENSOR_AXIS_SCORE_CALIBRATION);
+    if (score_calibration_filename.empty()) {
+      return CreateStatusWithPayload(
+          StatusCode::kNotFound,
+          "Found ScoreCalibrationOptions but missing required associated "
+          "parameters file with type TENSOR_AXIS_SCORE_CALIBRATION.",
+          TfLiteSupportStatus::kMetadataAssociatedFileNotFoundError);
+    }
+    ASSIGN_OR_RETURN(
+        absl::string_view score_calibration_file,
+        metadata_extractor.GetAssociatedFile(score_calibration_filename));
+    ASSIGN_OR_RETURN(SigmoidCalibrationParameters sigmoid_params,
+                     BuildSigmoidCalibrationParams(
+                         *score_calibration_process_unit
+                              ->options_as_ScoreCalibrationOptions(),
+                         score_calibration_file, head.label_map_items));
+    head.calibration_params = sigmoid_params;
+  }
+
+  return head;
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/core/classification_head.h b/tensorflow_lite_support/cc/task/vision/core/classification_head.h
new file mode 100644
index 0000000..07cd8b9
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/classification_head.h
@@ -0,0 +1,110 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_CLASSIFICATION_HEAD_ITEM_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_CLASSIFICATION_HEAD_ITEM_H_
+
+#include <string>
+#include <vector>
+
+#include "absl/memory/memory.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/vision/core/label_map_item.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/score_calibration.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// A single classifier head for an image classifier model, associated with a
+// corresponding output tensor.
+struct ClassificationHead {
+  ClassificationHead() : score_threshold(0) {}
+
+  explicit ClassificationHead(
+      const std::vector<tflite::task::vision::LabelMapItem>&& label_map_items)
+      : label_map_items(label_map_items), score_threshold(0) {}
+
+  // An optional name that usually indicates what this set of classes represent,
+  // e.g. "flowers".
+  std::string name;
+  // The label map representing the list of supported classes, aka labels.
+  //
+  // This must be in direct correspondence with the associated output tensor,
+  // i.e.:
+  //
+  // - The number of classes must match with the dimension of the corresponding
+  // output tensor,
+  // - The i-th item in the label map is assumed to correspond to the i-th
+  // output value in the output tensor.
+  //
+  // This requires to put in place dedicated sanity checks before running
+  // inference.
+  std::vector<tflite::task::vision::LabelMapItem> label_map_items;
+  // Recommended score threshold typically in [0,1[. Classification results with
+  // a score below this value are considered low-confidence and should be
+  // rejected from returned results.
+  float score_threshold;
+  // Optional score calibration parameters (one set of parameters per class in
+  // the label map). This is primarily meant for multi-label classifiers made of
+  // independent sigmoids.
+  //
+  // Such parameters are usually tuned so that calibrated scores can be compared
+  // to a default threshold common to all classes to achieve a given amount of
+  // precision.
+  //
+  // Example: 60% precision for threshold = 0.5.
+  absl::optional<tflite::task::vision::SigmoidCalibrationParameters>
+      calibration_params;
+};
+
+// Builds a classification head using the provided metadata extractor, for the
+// given output tensor metadata. Returns an error in case the head cannot be
+// built (e.g. missing associated file for score calibration parameters).
+//
+// Optionally it is possible to specify which locale should be used (e.g. "en")
+// to fill the label map display names, if any, and provided the corresponding
+// associated file is present in the metadata. If no locale is specified, or if
+// there is no associated file for the provided locale, display names are just
+// left empty and no error is returned.
+//
+// E.g. (metatada displayed in JSON format below):
+//
+// ...
+// "associated_files": [
+//  {
+//    "name": "labels.txt",
+//    "type": "TENSOR_AXIS_LABELS"
+//  },
+//  {
+//    "name": "labels-en.txt",
+//    "type": "TENSOR_AXIS_LABELS",
+//    "locale": "en"
+//  },
+// ...
+//
+// See metadata schema TENSOR_AXIS_LABELS for more details.
+tflite::support::StatusOr<ClassificationHead> BuildClassificationHead(
+    const tflite::metadata::ModelMetadataExtractor& metadata_extractor,
+    const tflite::TensorMetadata& output_tensor_metadata,
+    absl::string_view display_names_locale = absl::string_view());
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_CLASSIFICATION_HEAD_ITEM_H_
diff --git a/tensorflow_lite_support/cc/task/vision/core/frame_buffer.cc b/tensorflow_lite_support/cc/task/vision/core/frame_buffer.cc
new file mode 100644
index 0000000..02658cd
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/frame_buffer.cc
@@ -0,0 +1,179 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+using ::tflite::support::StatusOr;
+
+namespace {
+
+// Returns whether the input `format` is a supported YUV format.
+bool IsSupportedYuvFormat(FrameBuffer::Format format) {
+  return format == FrameBuffer::Format::kNV21 ||
+         format == FrameBuffer::Format::kNV12 ||
+         format == FrameBuffer::Format::kYV12 ||
+         format == FrameBuffer::Format::kYV21;
+}
+
+// Returns supported 1-plane FrameBuffer in YuvData structure.
+StatusOr<FrameBuffer::YuvData> GetYuvDataFromOnePlaneFrameBuffer(
+    const FrameBuffer& source) {
+  if (!IsSupportedYuvFormat(source.format())) {
+    return absl::InvalidArgumentError(
+        "The source FrameBuffer format is not part of YUV420 family.");
+  }
+
+  FrameBuffer::YuvData result;
+  const int y_buffer_size =
+      source.plane(0).stride.row_stride_bytes * source.dimension().height;
+  const int uv_buffer_size =
+      ((source.plane(0).stride.row_stride_bytes + 1) / 2) *
+      ((source.dimension().height + 1) / 2);
+  result.y_buffer = source.plane(0).buffer;
+  result.y_row_stride = source.plane(0).stride.row_stride_bytes;
+  result.uv_row_stride = result.y_row_stride;
+
+  if (source.format() == FrameBuffer::Format::kNV21) {
+    result.v_buffer = result.y_buffer + y_buffer_size;
+    result.u_buffer = result.v_buffer + 1;
+    result.uv_pixel_stride = 2;
+    // If y_row_stride equals to the frame width and is an odd value,
+    // uv_row_stride = y_row_stride + 1, otherwise uv_row_stride = y_row_stride.
+    if (result.y_row_stride == source.dimension().width &&
+        result.y_row_stride % 2 == 1) {
+      result.uv_row_stride = (result.y_row_stride + 1) / 2 * 2;
+    }
+  } else if (source.format() == FrameBuffer::Format::kNV12) {
+    result.u_buffer = result.y_buffer + y_buffer_size;
+    result.v_buffer = result.u_buffer + 1;
+    result.uv_pixel_stride = 2;
+    // If y_row_stride equals to the frame width and is an odd value,
+    // uv_row_stride = y_row_stride + 1, otherwise uv_row_stride = y_row_stride.
+    if (result.y_row_stride == source.dimension().width &&
+        result.y_row_stride % 2 == 1) {
+      result.uv_row_stride = (result.y_row_stride + 1) / 2 * 2;
+    }
+  } else if (source.format() == FrameBuffer::Format::kYV21) {
+    result.u_buffer = result.y_buffer + y_buffer_size;
+    result.v_buffer = result.u_buffer + uv_buffer_size;
+    result.uv_pixel_stride = 1;
+    result.uv_row_stride = (result.y_row_stride + 1) / 2;
+  } else if (source.format() == FrameBuffer::Format::kYV12) {
+    result.v_buffer = result.y_buffer + y_buffer_size;
+    result.u_buffer = result.v_buffer + uv_buffer_size;
+    result.uv_pixel_stride = 1;
+    result.uv_row_stride = (result.y_row_stride + 1) / 2;
+  }
+  return result;
+}
+
+// Returns supported 2-plane FrameBuffer in YuvData structure.
+StatusOr<FrameBuffer::YuvData> GetYuvDataFromTwoPlaneFrameBuffer(
+    const FrameBuffer& source) {
+  if (source.format() != FrameBuffer::Format::kNV12 &&
+      source.format() != FrameBuffer::Format::kNV21) {
+    return absl::InvalidArgumentError("Unsupported YUV planar format.");
+  }
+
+  FrameBuffer::YuvData result;
+  // Y plane
+  result.y_buffer = source.plane(0).buffer;
+  // All plane strides
+  result.y_row_stride = source.plane(0).stride.row_stride_bytes;
+  result.uv_row_stride = source.plane(1).stride.row_stride_bytes;
+  result.uv_pixel_stride = 2;
+
+  if (source.format() == FrameBuffer::Format::kNV12) {
+    // Y and UV interleaved format
+    result.u_buffer = source.plane(1).buffer;
+    result.v_buffer = result.u_buffer + 1;
+  } else {
+    // Y and VU interleaved format
+    result.v_buffer = source.plane(1).buffer;
+    result.u_buffer = result.v_buffer + 1;
+  }
+  return result;
+}
+
+// Returns supported 3-plane FrameBuffer in YuvData structure. Note that NV21
+// and NV12 are included in the supported Yuv formats. Technically, NV21 and
+// NV12 should not be described by the 3-plane format. Historically, NV21 is
+// used loosely such that it can also be used to describe YV21 format. For
+// backwards compatibility, FrameBuffer supports NV21/NV12 with 3-plane format
+// but such usage is discouraged
+StatusOr<FrameBuffer::YuvData> GetYuvDataFromThreePlaneFrameBuffer(
+    const FrameBuffer& source) {
+  if (!IsSupportedYuvFormat(source.format())) {
+    return absl::InvalidArgumentError(
+        "The source FrameBuffer format is not part of YUV420 family.");
+  }
+
+  if (source.plane(1).stride.row_stride_bytes !=
+          source.plane(2).stride.row_stride_bytes ||
+      source.plane(1).stride.pixel_stride_bytes !=
+          source.plane(2).stride.pixel_stride_bytes) {
+    return absl::InternalError("Unsupported YUV planar format.");
+  }
+  FrameBuffer::YuvData result;
+  if (source.format() == FrameBuffer::Format::kNV21 ||
+      source.format() == FrameBuffer::Format::kYV12) {
+    // Y follow by VU order. The VU chroma planes can be interleaved or
+    // planar.
+    result.y_buffer = source.plane(0).buffer;
+    result.v_buffer = source.plane(1).buffer;
+    result.u_buffer = source.plane(2).buffer;
+    result.y_row_stride = source.plane(0).stride.row_stride_bytes;
+    result.uv_row_stride = source.plane(1).stride.row_stride_bytes;
+    result.uv_pixel_stride = source.plane(1).stride.pixel_stride_bytes;
+  } else {
+    // Y follow by UV order. The UV chroma planes can be interleaved or
+    // planar.
+    result.y_buffer = source.plane(0).buffer;
+    result.u_buffer = source.plane(1).buffer;
+    result.v_buffer = source.plane(2).buffer;
+    result.y_row_stride = source.plane(0).stride.row_stride_bytes;
+    result.uv_row_stride = source.plane(1).stride.row_stride_bytes;
+    result.uv_pixel_stride = source.plane(1).stride.pixel_stride_bytes;
+  }
+  return result;
+}
+
+}  // namespace
+
+StatusOr<FrameBuffer::YuvData> FrameBuffer::GetYuvDataFromFrameBuffer(
+    const FrameBuffer& source) {
+  if (!IsSupportedYuvFormat(source.format())) {
+    return absl::InvalidArgumentError(
+        "The source FrameBuffer format is not part of YUV420 family.");
+  }
+
+  if (source.plane_count() == 1) {
+    return GetYuvDataFromOnePlaneFrameBuffer(source);
+  } else if (source.plane_count() == 2) {
+    return GetYuvDataFromTwoPlaneFrameBuffer(source);
+  } else if (source.plane_count() == 3) {
+    return GetYuvDataFromThreePlaneFrameBuffer(source);
+  }
+  return absl::InvalidArgumentError(
+      "The source FrameBuffer must be consisted by 1, 2, or 3 planes");
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/core/frame_buffer.h b/tensorflow_lite_support/cc/task/vision/core/frame_buffer.h
new file mode 100644
index 0000000..31589f3
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/frame_buffer.h
@@ -0,0 +1,296 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_FRAME_BUFFER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_FRAME_BUFFER_H_
+
+#include <map>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/memory/memory.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_cat.h"
+#include "absl/time/clock.h"
+#include "absl/time/time.h"
+#include "absl/types/any.h"
+#include "absl/types/optional.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// A `FrameBuffer` provides a view into the provided backing buffer (e.g. camera
+// frame or still image) with buffer format information. FrameBuffer doesn't
+// take ownership of the provided backing buffer. The caller is responsible to
+// manage the backing buffer lifecycle for the lifetime of the FrameBuffer.
+//
+// FrameBuffer also provides a tagging system to allow the client of FrameBuffer
+// to attach arbitrary tags to an instance. The tagging system is meant for
+// small set of metadata. FrameBuffer does not use the tags in anyway. The
+// uniqueness of the tag is only guarded by the uniqueness of the key.
+// The tag is useful when the uniqueness of a FrameBuffer can not be determined
+// by its associated metadata. For example, there are two FrameBuffer instances
+// with the same metadata (size dimension, orientation, format, etc) but one is
+// generated through cropping of Frame A and another is generated by resizing of
+// Frame A. The client can tag one of the generated FrameBuffer to distinguish
+// the difference.
+//
+// Examples:
+//
+// // Create an metadata instance with no backing buffer.
+// auto buffer = FrameBuffer::Create(/*planes=*/{}, dimension, kRGBA,
+//                                   KTopLeft);
+//
+// // Create an RGBA instance with backing buffer on single plane.
+// FrameBuffer::Plane plane = {rgba_buffer, /*stride=*/{dimension.width * 4,
+// 4}}; auto buffer = FrameBuffer::Create({plane}, dimension, kRGBA, kTopLeft);
+//
+// // Create an YUV instance with planar backing buffer.
+// FrameBuffer::Plane y_plane = {y_buffer, /*stride=*/{dimension.width , 1}};
+// FrameBuffer::Plane uv_plane = {u_buffer, /*stride=*/{dimension.width, 2}};
+// auto buffer = FrameBuffer::Create({y_plane, uv_plane}, dimension, kNV21,
+//                                   kLeftTop);
+//
+// // Add / retrieve tags from a FrameBuffer instance.
+// buffer.InsertTag("my_special_key", 1);
+// buffer.GetTag("my_special_key");
+//
+class FrameBuffer {
+ public:
+  // Colorspace formats.
+  enum class Format { kRGBA, kRGB, kNV12, kNV21, kYV12, kYV21, kGRAY };
+
+  // Stride information.
+  struct Stride {
+    // The row stride in bytes. This is the distance between the start pixels of
+    // two consecutive rows in the image.
+    int row_stride_bytes;
+    // This is the distance between two consecutive pixel values in a row of
+    // pixels in bytes. It may be larger than the size of a single pixel to
+    // account for interleaved image data or padded formats.
+    int pixel_stride_bytes;
+  };
+
+  // YUV data structure.
+  struct YuvData {
+    const uint8* y_buffer;
+    const uint8* u_buffer;
+    const uint8* v_buffer;
+    // Y buffer row stride in bytes.
+    int y_row_stride;
+    // U/V buffer row stride in bytes.
+    int uv_row_stride;
+    // U/V pixel stride in bytes. This is the distance between two consecutive
+    // u/v pixel values in a row.
+    int uv_pixel_stride;
+  };
+
+  // FrameBuffer content orientation follows EXIF specification. The name of
+  // each enum value defines the position of the 0th row and the 0th column of
+  // the image content. See http://jpegclub.org/exif_orientation.html for
+  // details.
+  enum class Orientation {
+    kTopLeft = 1,
+    kTopRight = 2,
+    kBottomRight = 3,
+    kBottomLeft = 4,
+    kLeftTop = 5,
+    kRightTop = 6,
+    kRightBottom = 7,
+    kLeftBottom = 8
+  };
+
+  // Plane encapsulates buffer and stride information.
+  struct Plane {
+    const uint8* buffer;
+    Stride stride;
+  };
+
+  // Dimension information for the whole frame or a cropped portion of it.
+  struct Dimension {
+    // The width dimension in pixel unit.
+    int width;
+    // The height dimension in pixel unit.
+    int height;
+
+    bool operator==(const Dimension& other) const {
+      return width == other.width && height == other.height;
+    }
+
+    bool operator!=(const Dimension& other) const {
+      return width != other.width || height != other.height;
+    }
+
+    bool operator>=(const Dimension& other) const {
+      return width >= other.width && height >= other.height;
+    }
+
+    bool operator<=(const Dimension& other) const {
+      return width <= other.width && height <= other.height;
+    }
+
+    // Swaps width and height.
+    void Swap() {
+      using std::swap;
+      swap(width, height);
+    }
+
+    // Returns area represented by width * height.
+    int Size() const { return width * height; }
+  };
+
+  // Factory method for creating a FrameBuffer object from row-major backing
+  // buffers. In a streaming use case (e.g continuous camera stream), the
+  // timestamp can be used as an ID to identify a frame.
+  static std::unique_ptr<FrameBuffer> Create(const std::vector<Plane>& planes,
+                                             Dimension dimension, Format format,
+                                             Orientation orientation,
+                                             absl::Time timestamp) {
+    return absl::make_unique<FrameBuffer>(planes, dimension, format,
+                                          orientation, timestamp);
+  }
+
+  // Factory method for creating a FrameBuffer object from row-major movable
+  // backing buffers. In a streaming use case (e.g continuous camera stream),
+  // the timestamp can be used as an ID to identify a frame.
+  static std::unique_ptr<FrameBuffer> Create(std::vector<Plane>&& planes,
+                                             Dimension dimension, Format format,
+                                             Orientation orientation,
+                                             absl::Time timestamp) {
+    return absl::make_unique<FrameBuffer>(std::move(planes), dimension, format,
+                                          orientation, timestamp);
+  }
+
+  // Factory method for creating a FrameBuffer object from row-major backing
+  // buffers. By default this method set the timestamp to now. This method is
+  // more suitable for processing use case that does not need to re-identify
+  // this buffer.
+  static std::unique_ptr<FrameBuffer> Create(const std::vector<Plane>& planes,
+                                             Dimension dimension, Format format,
+                                             Orientation orientation) {
+    return absl::make_unique<FrameBuffer>(planes, dimension, format,
+                                          orientation, absl::Now());
+  }
+
+  // Factory method for creating a FrameBuffer object from movable row-major
+  // backing buffers. By default this method set the timestamp to now. This
+  // method is more suitable for processing use case that does not need to
+  // re-identify this buffer.
+  static std::unique_ptr<FrameBuffer> Create(std::vector<Plane>&& planes,
+                                             Dimension dimension, Format format,
+                                             Orientation orientation) {
+    return absl::make_unique<FrameBuffer>(std::move(planes), dimension, format,
+                                          orientation, absl::Now());
+  }
+
+  // Returns YuvData which contains the Y, U, and V buffer and their
+  // stride info from the input `source` FrameBuffer which is in the YUV family
+  // formats (e.g NV12, NV21, YV12, and YV21).
+  static tflite::support::StatusOr<YuvData> GetYuvDataFromFrameBuffer(
+      const FrameBuffer& source);
+
+  // Builds a FrameBuffer object from a row-major backing buffer.
+  //
+  // The FrameBuffer does not take ownership of the backing buffer. The backing
+  // buffer is read-only and the caller is responsible for maintaining the
+  // backing buffer lifecycle for the lifetime of FrameBuffer.
+  FrameBuffer(const std::vector<Plane>& planes, Dimension dimension,
+              Format format, Orientation orientation, absl::Time timestamp)
+      : planes_(planes),
+        dimension_(dimension),
+        format_(format),
+        orientation_(orientation),
+        timestamp_(timestamp) {}
+
+  // Builds a FrameBuffer object from a movable row-major backing buffer.
+  //
+  // The FrameBuffer does not take ownership of the backing buffer. The backing
+  // buffer is read-only and the caller is responsible for maintaining the
+  // backing buffer lifecycle for the lifetime of FrameBuffer.
+  FrameBuffer(std::vector<Plane>&& planes, Dimension dimension, Format format,
+              Orientation orientation, absl::Time timestamp)
+      : planes_(std::move(planes)),
+        dimension_(dimension),
+        format_(format),
+        orientation_(orientation),
+        timestamp_(timestamp) {}
+
+  // Returns number of planes.
+  const int plane_count() const { return planes_.size(); }
+
+  // Returns plane indexed by the input `index`.
+  const Plane plane(int index) const {
+    if (index > -1 && index < planes_.size()) {
+      return planes_[index];
+    }
+    return {};
+  }
+
+  // Returns the tag associated to the tag_key.
+  absl::any GetTag(const std::string& tag_key) const {
+    auto iter = tags_.find(tag_key);
+    if (iter != tags_.end()) {
+      return iter->second;
+    }
+    return absl::any();
+  }
+
+  // Inserts or updates the tags map with key value pair (tag_key, tag_value).
+  void InsertOrUpdateTag(const std::string& tag_key, absl::any tag_value) {
+    tags_[tag_key] = std::move(tag_value);
+  }
+
+  // Inserts the key value pair (tag_key, tag_value) into tags map. If the
+  // tag_key already exists, an internal error will return.
+  absl::Status InsertTag(const std::string& tag_key, absl::any tag_value) {
+    auto iter = tags_.emplace(tag_key, tag_value);
+    if (iter.second) {
+      return absl::OkStatus();
+    }
+    return absl::InternalError(absl::StrCat(
+        "tag_key already exists in tags.tag_key was not inserted: ", tag_key));
+  }
+
+  // Returns FrameBuffer dimension.
+  const Dimension dimension() const { return dimension_; }
+
+  // Returns FrameBuffer format.
+  const Format format() const { return format_; }
+
+  // Returns FrameBuffer orientation.
+  const Orientation orientation() const { return orientation_; }
+
+  // Returns FrameBuffer timestamp.
+  const absl::Time timestamp() const { return timestamp_; }
+
+ private:
+  std::vector<Plane> planes_;
+  std::map<std::string, absl::any> tags_;
+  Dimension dimension_;
+  Format format_;
+  Orientation orientation_;
+  absl::Time timestamp_;
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_FRAME_BUFFER_H_
diff --git a/tensorflow_lite_support/cc/task/vision/core/label_map_item.cc b/tensorflow_lite_support/cc/task/vision/core/label_map_item.cc
new file mode 100644
index 0000000..75b1fc6
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/label_map_item.cc
@@ -0,0 +1,128 @@
+
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/cc/task/vision/core/label_map_item.h"
+
+#include "absl/strings/str_format.h"
+#include "absl/strings/str_split.h"
+#include "tensorflow_lite_support/cc/common.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+using ::absl::StatusCode;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+
+StatusOr<std::vector<LabelMapItem>> BuildLabelMapFromFiles(
+    absl::string_view labels_file, absl::string_view display_names_file) {
+  if (labels_file.empty()) {
+    return CreateStatusWithPayload(StatusCode::kInvalidArgument,
+                                   "Expected non-empty labels file.",
+                                   TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  std::vector<absl::string_view> labels = absl::StrSplit(labels_file, '\n');
+  // In most cases, there is an empty line (i.e. newline character) at the end
+  // of the file that needs to be ignored. In such a situation, StrSplit() will
+  // produce a vector with an empty string as final element. Also note that in
+  // case `labels_file` is entirely empty, StrSplit() will produce a vector with
+  // one single empty substring, so there's no out-of-range risk here.
+  if (labels[labels.size() - 1].empty()) {
+    labels.pop_back();
+  }
+
+  std::vector<LabelMapItem> label_map_items;
+  label_map_items.reserve(labels.size());
+  for (int i = 0; i < labels.size(); ++i) {
+    label_map_items.emplace_back(LabelMapItem{.name = std::string(labels[i])});
+  }
+
+  if (!display_names_file.empty()) {
+    std::vector<std::string> display_names =
+        absl::StrSplit(display_names_file, '\n');
+    // In most cases, there is an empty line (i.e. newline character) at the end
+    // of the file that needs to be ignored. See above.
+    if (display_names[display_names.size() - 1].empty()) {
+      display_names.pop_back();
+    }
+    if (display_names.size() != labels.size()) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat(
+              "Mismatch between number of labels (%d) and display names (%d).",
+              labels.size(), display_names.size()),
+          TfLiteSupportStatus::kMetadataNumLabelsMismatchError);
+    }
+    for (int i = 0; i < display_names.size(); ++i) {
+      label_map_items[i].display_name = display_names[i];
+    }
+  }
+  return label_map_items;
+}
+
+absl::Status LabelHierarchy::InitializeFromLabelMap(
+    std::vector<LabelMapItem> label_map_items) {
+  parents_map_.clear();
+  for (const LabelMapItem& label : label_map_items) {
+    for (const std::string& child_name : label.child_name) {
+      parents_map_[child_name].insert(label.name);
+    }
+  }
+  if (parents_map_.empty()) {
+    return CreateStatusWithPayload(StatusCode::kInvalidArgument,
+                                   "Input labelmap is not hierarchical: there "
+                                   "is no parent-child relationship.",
+                                   TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  return absl::OkStatus();
+}
+
+bool LabelHierarchy::HaveAncestorDescendantRelationship(
+    const std::string& ancestor_name,
+    const std::string& descendant_name) const {
+  absl::flat_hash_set<std::string> ancestors;
+  GetAncestors(descendant_name, &ancestors);
+  return ancestors.contains(ancestor_name);
+}
+
+absl::flat_hash_set<std::string> LabelHierarchy::GetParents(
+    const std::string& name) const {
+  absl::flat_hash_set<std::string> parents;
+  auto it = parents_map_.find(name);
+  if (it != parents_map_.end()) {
+    for (const std::string& parent_name : it->second) {
+      parents.insert(parent_name);
+    }
+  }
+  return parents;
+}
+
+void LabelHierarchy::GetAncestors(
+    const std::string& name,
+    absl::flat_hash_set<std::string>* ancestors) const {
+  const absl::flat_hash_set<std::string> parents = GetParents(name);
+  for (const std::string& parent_name : parents) {
+    auto it = ancestors->insert(parent_name);
+    if (it.second) {
+      GetAncestors(parent_name, ancestors);
+    }
+  }
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/core/label_map_item.h b/tensorflow_lite_support/cc/task/vision/core/label_map_item.h
new file mode 100644
index 0000000..3ac9a00
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/core/label_map_item.h
@@ -0,0 +1,95 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_LABEL_MAP_ITEM_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_LABEL_MAP_ITEM_H_
+
+#include <string>
+#include <vector>
+
+#include "absl/container/flat_hash_map.h"
+#include "absl/container/flat_hash_set.h"
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Structure mapping a numerical class index output to a Knowledge Graph entity
+// ID or any other string label representing this class. Optionally it is
+// possible to specify an additional display name (in a given language) which is
+// typically used for display purposes.
+struct LabelMapItem {
+  // E.g. name = "/m/02xwb"
+  std::string name;
+  // E.g. display_name = "Fruit"
+  std::string display_name;
+  // Optional list of children (e.g. subcategories) used to represent a
+  // hierarchy.
+  std::vector<std::string> child_name;
+};
+
+// Builds a label map from labels and (optional) display names file contents,
+// both expected to contain one label per line. Those are typically obtained
+// from TFLite Model Metadata TENSOR_AXIS_LABELS or TENSOR_VALUE_LABELS
+// associated files.
+// Returns an error e.g. if there's a mismatch between the number of labels and
+// display names.
+tflite::support::StatusOr<std::vector<LabelMapItem>> BuildLabelMapFromFiles(
+    absl::string_view labels_file, absl::string_view display_names_file);
+
+// A class that represents a hierarchy of labels as specified in a label map.
+//
+// For example, it is useful to determine if one label is a descendant of
+// another label or not. This can be used to implement labels pruning based on
+// hierarchy, e.g. if both "fruit" and "banana" have been inferred by a given
+// classifier model prune "fruit" from the final results as "banana" is a more
+// fine-grained descendant.
+class LabelHierarchy {
+ public:
+  LabelHierarchy() = default;
+
+  // Initializes the hierarchy of labels from a given label map vector. Returns
+  // an error status in case of failure, typically if the input label map does
+  // not contain any hierarchical relations between labels.
+  absl::Status InitializeFromLabelMap(
+      std::vector<LabelMapItem> label_map_items);
+
+  // Returns true if `descendant_name` is a descendant of `ancestor_name` in the
+  // hierarchy of labels. Invalid names, i.e. names which do not exist in the
+  // label map used at initialization time, are ignored.
+  bool HaveAncestorDescendantRelationship(
+      const std::string& ancestor_name,
+      const std::string& descendant_name) const;
+
+ private:
+  // Retrieve and return all parent names, if any, for the input label name.
+  absl::flat_hash_set<std::string> GetParents(const std::string& name) const;
+
+  // Retrieve all ancestor names, if any, for the input label name.
+  void GetAncestors(const std::string& name,
+                    absl::flat_hash_set<std::string>* ancestors) const;
+
+  // Label name (key) to parent names (value) direct mapping.
+  absl::flat_hash_map<std::string, absl::flat_hash_set<std::string>>
+      parents_map_;
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_CORE_LABEL_MAP_ITEM_H_
diff --git a/tensorflow_lite_support/cc/task/vision/image_classifier.cc b/tensorflow_lite_support/cc/task/vision/image_classifier.cc
new file mode 100644
index 0000000..378797b
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/image_classifier.cc
@@ -0,0 +1,572 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/vision/image_classifier.h"
+
+#include "absl/algorithm/container.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/string_view.h"
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/interpreter.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/core/task_api_factory.h"
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+#include "tensorflow_lite_support/cc/task/vision/core/label_map_item.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+namespace {
+
+using ::absl::StatusCode;
+using ::tflite::metadata::ModelMetadataExtractor;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+using ::tflite::task::core::AssertAndReturnTypedTensor;
+using ::tflite::task::core::TaskAPIFactory;
+using ::tflite::task::core::TfLiteEngine;
+
+// Default score value used as a fallback for classes that (1) have no score
+// calibration data or (2) have a very low confident uncalibrated score, i.e.
+// lower than the `min_uncalibrated_score` threshold.
+//
+// (1) This happens when the ScoreCalibration does not cover all the classes
+// listed in the label map. This can be used to enforce the blacklisting of
+// given classes so that they are never returned.
+//
+// (2) This is an optional threshold provided part of the calibration data. It
+// is used to mitigate false alarms on some classes.
+//
+// In both cases, a class that gets assigned a score of -1 is never returned as
+// it gets discarded by the `score_threshold` check (see post-processing logic).
+constexpr float kDefaultCalibratedScore = -1.0f;
+
+// Calibrated scores should be in the [0, 1] range, otherwise an error is
+// returned at post-processing time.
+constexpr float kMinCalibratedScore = 0.0f;
+constexpr float kMaxCalibratedScore = 1.0f;
+
+}  // namespace
+
+/* static */
+StatusOr<std::unique_ptr<ImageClassifier>> ImageClassifier::CreateFromOptions(
+    const ImageClassifierOptions& options,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  RETURN_IF_ERROR(SanityCheckOptions(options));
+
+  // Copy options to ensure the ExternalFile outlives the constructed object.
+  auto options_copy = absl::make_unique<ImageClassifierOptions>(options);
+
+  ASSIGN_OR_RETURN(auto image_classifier,
+                   TaskAPIFactory::CreateFromExternalFileProto<ImageClassifier>(
+                       &options_copy->model_file_with_metadata(),
+                       std::move(resolver), options_copy->num_threads()));
+
+  RETURN_IF_ERROR(image_classifier->Init(std::move(options_copy)));
+
+  return image_classifier;
+}
+
+/* static */
+absl::Status ImageClassifier::SanityCheckOptions(
+    const ImageClassifierOptions& options) {
+  if (!options.has_model_file_with_metadata()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Missing mandatory `model_file_with_metadata` field",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.max_results() == 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Invalid `max_results` option: value must be != 0",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.score_threshold() < 0 || options.score_threshold() >= 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "`score_threshold` out of range: %f. Valid range is [0,1[.",
+            options.score_threshold()),
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.class_name_whitelist_size() > 0 &&
+      options.class_name_blacklist_size() > 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "`class_name_whitelist` and `class_name_blacklist` are mutually "
+        "exclusive options.",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.num_threads() == 0 || options.num_threads() < -1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "`num_threads` must be greater than 0 or equal to -1.",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  return absl::OkStatus();
+}
+
+absl::Status ImageClassifier::Init(
+    std::unique_ptr<ImageClassifierOptions> options) {
+  // Set options.
+  options_ = std::move(options);
+
+  // Perform pre-initialization actions (by default, sets the process engine for
+  // image pre-processing to kLibyuv as a sane default).
+  RETURN_IF_ERROR(PreInit());
+
+  // Sanity check and set inputs and outputs.
+  RETURN_IF_ERROR(CheckAndSetInputs());
+  RETURN_IF_ERROR(CheckAndSetOutputs());
+
+  // Initialize class whitelisting/blacklisting, if any.
+  RETURN_IF_ERROR(CheckAndSetClassNameSet());
+
+  // Perform final initialization (by default, initialize score calibration
+  // parameters, if any).
+  RETURN_IF_ERROR(PostInit());
+
+  return absl::OkStatus();
+}
+
+absl::Status ImageClassifier::PreInit() {
+  SetProcessEngine(FrameBufferUtils::ProcessEngine::kLibyuv);
+  return absl::OkStatus();
+}
+
+absl::Status ImageClassifier::PostInit() { return InitScoreCalibrations(); }
+
+absl::Status ImageClassifier::CheckAndSetOutputs() {
+  num_outputs_ = TfLiteEngine::OutputCount(engine_->interpreter());
+
+  // Perform sanity checks and extract metadata.
+  const ModelMetadataExtractor* metadata_extractor =
+      engine_->metadata_extractor();
+
+  const flatbuffers::Vector<flatbuffers::Offset<tflite::TensorMetadata>>*
+      output_tensor_metadata = metadata_extractor->GetOutputTensorMetadata();
+
+  // Loop over output tensors metadata, if any.
+  // Note: models with no output tensor metadata at all are supported.
+  if (output_tensor_metadata != nullptr) {
+    int num_output_tensors = output_tensor_metadata->size();
+
+    if (num_outputs_ != num_output_tensors) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Mismatch between number of output tensors (%d) and "
+                          "output tensors "
+                          "metadata (%d).",
+                          num_outputs_, num_output_tensors),
+          TfLiteSupportStatus::kMetadataInconsistencyError);
+    }
+
+    for (int i = 0; i < num_output_tensors; ++i) {
+      const tflite::TensorMetadata* output_tensor =
+          output_tensor_metadata->Get(i);
+
+      ASSIGN_OR_RETURN(
+          ClassificationHead head,
+          BuildClassificationHead(*metadata_extractor, *output_tensor,
+                                  options_->display_names_locale()));
+
+      classification_heads_.emplace_back(std::move(head));
+    }
+  }
+
+  // If classifier heads are not set, build default ones based on model
+  // introspection. This happens if a model with partial or no metadata was
+  // provided through the `model_file_with_metadata` options field.
+  if (classification_heads_.empty()) {
+    classification_heads_.reserve(num_outputs_);
+    for (int output_index = 0; output_index < num_outputs_; ++output_index) {
+      classification_heads_.emplace_back(ClassificationHead{});
+    }
+  }
+
+  if (num_outputs_ != classification_heads_.size()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Got %d classifier head(s), expected %d according to "
+                        "the label map.",
+                        num_outputs_, classification_heads_.size()),
+        TfLiteSupportStatus::kMetadataInconsistencyError);
+  }
+
+  int num_quantized_outputs = 0;
+  for (int i = 0; i < num_outputs_; ++i) {
+    const TfLiteTensor* output_tensor =
+        TfLiteEngine::GetOutput(engine_->interpreter(), i);
+    const int num_dimensions = output_tensor->dims->size;
+    if (num_dimensions == 4) {
+      if (output_tensor->dims->data[1] != 1 ||
+          output_tensor->dims->data[2] != 1) {
+        return CreateStatusWithPayload(
+            StatusCode::kInvalidArgument,
+            absl::StrFormat("Unexpected WxH sizes for output index %d: got "
+                            "%dx%d, expected 1x1.",
+                            i, output_tensor->dims->data[2],
+                            output_tensor->dims->data[1]),
+            TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+      }
+    } else if (num_dimensions != 2) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat(
+              "Unexpected number of dimensions for output index %d: got %dD, "
+              "expected either 2D (BxN with B=1) or 4D (BxHxWxN with B=1, W=1, "
+              "H=1).",
+              i, num_dimensions),
+          TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+    }
+    if (output_tensor->dims->data[0] != 1) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("The output array is expected to have a batch size "
+                          "of 1. Got %d for output index %d.",
+                          output_tensor->dims->data[0], i),
+          TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+    }
+    int num_classes = output_tensor->dims->data[num_dimensions - 1];
+    // If label map is not set, build a default one based on model
+    // introspection. This happens if a model with partial or no metadata was
+    // provided through the `model_file_with_metadata` options field.
+    if (classification_heads_[i].label_map_items.empty()) {
+      classification_heads_[i].label_map_items.reserve(num_classes);
+      for (int class_index = 0; class_index < num_classes; ++class_index) {
+        classification_heads_[i].label_map_items.emplace_back(LabelMapItem{});
+      }
+    }
+    int num_label_map_items = classification_heads_[i].label_map_items.size();
+    if (num_classes != num_label_map_items) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Got %d class(es) for output index %d, expected %d "
+                          "according to the label map.",
+                          output_tensor->dims->data[num_dimensions - 1], i,
+                          num_label_map_items),
+          TfLiteSupportStatus::kMetadataInconsistencyError);
+    }
+    if (output_tensor->type == kTfLiteUInt8) {
+      num_quantized_outputs++;
+    } else if (output_tensor->type != kTfLiteFloat32) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Type mismatch for output tensor %s. Requested one "
+                          "of these types: "
+                          "kTfLiteUint8/kTfLiteFloat32, got %s.",
+                          output_tensor->name,
+                          TfLiteTypeGetName(output_tensor->type)),
+          TfLiteSupportStatus::kInvalidOutputTensorTypeError);
+    }
+  }
+
+  if (num_quantized_outputs > 0 && num_quantized_outputs != num_outputs_) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Got %d quantized output(s), expected %d (i.e. all "
+                        "provided outputs must be quantized).",
+                        num_quantized_outputs, num_outputs_),
+        TfLiteSupportStatus::kInvalidOutputTensorTypeError);
+  }
+  has_uint8_outputs_ = (num_quantized_outputs > 0);
+
+  return absl::OkStatus();
+}
+
+absl::Status ImageClassifier::CheckAndSetClassNameSet() {
+  // Exit early if no blacklist/whitelist.
+  if (options_->class_name_blacklist_size() == 0 &&
+      options_->class_name_whitelist_size() == 0) {
+    return absl::OkStatus();
+  }
+
+  // Before processing class names whitelist or blacklist from the input options
+  // create a set with _all_ known class names from the label map(s).
+  absl::flat_hash_set<std::string> all_class_names;
+  int head_index = 0;
+  for (const auto& head : classification_heads_) {
+    absl::flat_hash_set<std::string> head_class_names;
+    for (const auto& item : head.label_map_items) {
+      if (!item.name.empty()) {
+        head_class_names.insert(item.name);
+      }
+    }
+    if (head_class_names.empty()) {
+      std::string name = head.name;
+      if (name.empty()) {
+        name = absl::StrFormat("#%d", head_index);
+      }
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat(
+              "Using `class_name_whitelist` or `class_name_blacklist` "
+              "requires labels to be present but none was found for "
+              "classification head: %s",
+              name),
+          TfLiteSupportStatus::kMetadataMissingLabelsError);
+    }
+    all_class_names.insert(head_class_names.begin(), head_class_names.end());
+    head_index++;
+  }
+
+  class_name_set_.is_whitelist = options_->class_name_whitelist_size() > 0;
+  const auto& class_names = class_name_set_.is_whitelist
+                                ? options_->class_name_whitelist()
+                                : options_->class_name_blacklist();
+
+  // Note: duplicate or unknown classes are just ignored.
+  class_name_set_.values.clear();
+  for (const auto& class_name : class_names) {
+    if (!all_class_names.contains(class_name)) {
+      continue;
+    }
+    class_name_set_.values.insert(class_name);
+  }
+
+  if (class_name_set_.values.empty()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Invalid class names specified via `class_name_%s`: none match "
+            "with model labels.",
+            class_name_set_.is_whitelist ? "whitelist" : "blacklist"),
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+
+  return absl::OkStatus();
+}
+
+absl::Status ImageClassifier::InitScoreCalibrations() {
+  score_calibrations_.clear();
+  score_calibrations_.resize(classification_heads_.size());
+
+  for (int i = 0; i < classification_heads_.size(); ++i) {
+    if (!classification_heads_[i].calibration_params.has_value()) {
+      continue;
+    }
+
+    // Use a specific default score instead of the one specified by default in
+    // cc/task/vision/utils/score_calibration.h. See `kDefaultCalibratedScore`
+    // documentation for more details.
+    classification_heads_[i].calibration_params->default_score =
+        kDefaultCalibratedScore;
+
+    score_calibrations_[i] = absl::make_unique<ScoreCalibration>();
+    if (score_calibrations_[i] == nullptr) {
+      return CreateStatusWithPayload(
+          StatusCode::kInternal, "Could not create score calibration object.");
+    }
+
+    RETURN_IF_ERROR(score_calibrations_[i]->InitializeFromParameters(
+        classification_heads_[i].calibration_params.value()));
+  }
+
+  return absl::OkStatus();
+}
+
+StatusOr<ClassificationResult> ImageClassifier::Classify(
+    const FrameBuffer& frame_buffer) {
+  BoundingBox roi;
+  roi.set_width(frame_buffer.dimension().width);
+  roi.set_height(frame_buffer.dimension().height);
+  return Classify(frame_buffer, roi);
+}
+
+StatusOr<ClassificationResult> ImageClassifier::Classify(
+    const FrameBuffer& frame_buffer, const BoundingBox& roi) {
+  return InferWithFallback(frame_buffer, roi);
+}
+
+StatusOr<ClassificationResult> ImageClassifier::Postprocess(
+    const std::vector<const TfLiteTensor*>& output_tensors,
+    const FrameBuffer& /*frame_buffer*/, const BoundingBox& /*roi*/) {
+  if (output_tensors.size() != num_outputs_) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Expected %d output tensors, found %d", num_outputs_,
+                        output_tensors.size()));
+  }
+
+  ClassificationResult result;
+  std::vector<std::pair<int, float>> score_pairs;
+
+  for (int i = 0; i < num_outputs_; ++i) {
+    auto* classifications = result.add_classifications();
+    classifications->set_head_index(i);
+
+    const auto& head = classification_heads_[i];
+    score_pairs.clear();
+    score_pairs.reserve(head.label_map_items.size());
+
+    const TfLiteTensor* output_tensor = output_tensors[i];
+    if (has_uint8_outputs_) {
+      const uint8* output_data =
+          AssertAndReturnTypedTensor<uint8>(output_tensor);
+      for (int j = 0; j < head.label_map_items.size(); ++j) {
+        score_pairs.emplace_back(j, output_tensor->params.scale *
+                                        (static_cast<int>(output_data[j]) -
+                                         output_tensor->params.zero_point));
+      }
+    } else {
+      const float* output_data =
+          AssertAndReturnTypedTensor<float>(output_tensor);
+      for (int j = 0; j < head.label_map_items.size(); ++j) {
+        score_pairs.emplace_back(j, output_data[j]);
+      }
+    }
+
+    // Optional score calibration.
+    if (score_calibrations_[i] != nullptr) {
+      for (auto& score_pair : score_pairs) {
+        const std::string& class_name =
+            head.label_map_items[score_pair.first].name;
+        score_pair.second = score_calibrations_[i]->ComputeCalibratedScore(
+            class_name, score_pair.second);
+        if (score_pair.second > kMaxCalibratedScore) {
+          return CreateStatusWithPayload(
+              StatusCode::kInternal,
+              absl::StrFormat("calibrated score is too high: got %f, expected "
+                              "%f as maximum.",
+                              score_pair.second, kMaxCalibratedScore));
+        }
+        if (score_pair.second != kDefaultCalibratedScore &&
+            score_pair.second < kMinCalibratedScore) {
+          return CreateStatusWithPayload(
+              StatusCode::kInternal,
+              absl::StrFormat("calibrated score is too low: got %f, expected "
+                              "%f as minimum.",
+                              score_pair.second, kMinCalibratedScore));
+        }
+      }
+    }
+
+    int num_results =
+        options_->max_results() >= 0
+            ? std::min(static_cast<int>(head.label_map_items.size()),
+                       options_->max_results())
+            : head.label_map_items.size();
+    float score_threshold = options_->has_score_threshold()
+                                ? options_->score_threshold()
+                                : head.score_threshold;
+
+    if (class_name_set_.values.empty()) {
+      // Partially sort in descending order (higher score is better).
+      absl::c_partial_sort(
+          score_pairs, score_pairs.begin() + num_results,
+          [](const std::pair<int, float>& a, const std::pair<int, float>& b) {
+            return a.second > b.second;
+          });
+
+      for (int j = 0; j < num_results; ++j) {
+        float score = score_pairs[j].second;
+        if (score < score_threshold) {
+          break;
+        }
+        auto* cl = classifications->add_classes();
+        cl->set_index(score_pairs[j].first);
+        cl->set_score(score);
+      }
+    } else {
+      // Sort in descending order (higher score is better).
+      absl::c_sort(score_pairs, [](const std::pair<int, float>& a,
+                                   const std::pair<int, float>& b) {
+        return a.second > b.second;
+      });
+
+      for (int j = 0; j < head.label_map_items.size(); ++j) {
+        float score = score_pairs[j].second;
+        if (score < score_threshold ||
+            classifications->classes_size() >= num_results) {
+          break;
+        }
+
+        const int class_index = score_pairs[j].first;
+        const std::string& class_name = head.label_map_items[class_index].name;
+
+        bool class_name_found = class_name_set_.values.contains(class_name);
+
+        if ((!class_name_found && class_name_set_.is_whitelist) ||
+            (class_name_found && !class_name_set_.is_whitelist)) {
+          continue;
+        }
+
+        auto* cl = classifications->add_classes();
+        cl->set_index(class_index);
+        cl->set_score(score);
+      }
+    }
+  }
+
+  RETURN_IF_ERROR(FillResultsFromLabelMaps(&result));
+
+  return result;
+}
+
+absl::Status ImageClassifier::FillResultsFromLabelMaps(
+    ClassificationResult* result) {
+  for (int i = 0; i < result->classifications_size(); ++i) {
+    Classifications* classifications = result->mutable_classifications(i);
+    int head_index = classifications->head_index();
+    if (head_index < 0 || head_index >= classification_heads_.size()) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Invalid head index (%d) with respect to total "
+                          "number of classification heads (%d).",
+                          head_index, classification_heads_.size()),
+          TfLiteSupportStatus::kMetadataInconsistencyError);
+    }
+    const std::vector<LabelMapItem>& label_map_items =
+        classification_heads_[head_index].label_map_items;
+    for (int j = 0; j < classifications->classes_size(); ++j) {
+      Class* current_class = classifications->mutable_classes(j);
+      int current_class_index = current_class->index();
+      if (current_class_index < 0 ||
+          current_class_index >= label_map_items.size()) {
+        return CreateStatusWithPayload(
+            StatusCode::kInvalidArgument,
+            absl::StrFormat("Invalid class index (%d) with respect to label "
+                            "map size (%d) for head #%d.",
+                            current_class_index, label_map_items.size(),
+                            head_index),
+            TfLiteSupportStatus::kMetadataInconsistencyError);
+      }
+      const std::string& name = label_map_items[current_class_index].name;
+      if (!name.empty()) {
+        current_class->set_class_name(name);
+      }
+      const std::string& display_name =
+          label_map_items[current_class_index].display_name;
+      if (!display_name.empty()) {
+        current_class->set_display_name(display_name);
+      }
+    }
+  }
+  return absl::OkStatus();
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/image_classifier.h b/tensorflow_lite_support/cc/task/vision/image_classifier.h
new file mode 100644
index 0000000..edd9093
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/image_classifier.h
@@ -0,0 +1,182 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_CLASSIFIER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_CLASSIFIER_H_
+
+#include <memory>
+#include <vector>
+
+#include "absl/container/flat_hash_set.h"
+#include "absl/status/status.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+#include "tensorflow_lite_support/cc/task/vision/core/base_vision_task_api.h"
+#include "tensorflow_lite_support/cc/task/vision/core/classification_head.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/classifications_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/image_classifier_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/score_calibration.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Performs classification on images.
+//
+// The API expects a TFLite model with optional, but strongly recommended,
+// TFLite Model Metadata.
+//
+// Input tensor:
+//   (kTfLiteUInt8/kTfLiteFloat32)
+//    - image input of size `[batch x height x width x channels]`.
+//    - batch inference is not supported (`batch` is required to be 1).
+//    - only RGB inputs are supported (`channels` is required to be 3).
+//    - if type is kTfLiteFloat32, NormalizationOptions are required to be
+//      attached to the metadata for input normalization.
+// At least one output tensor with:
+//   (kTfLiteUInt8/kTfLiteFloat32)
+//    -  `N `classes and either 2 or 4 dimensions, i.e. `[1 x N]` or
+//       `[1 x 1 x 1 x N]`
+//    - optional (but recommended) label map(s) as AssociatedFile-s with type
+//      TENSOR_AXIS_LABELS, containing one label per line. The first such
+//      AssociatedFile (if any) is used to fill the `class_name` field of the
+//      results. The `display_name` field is filled from the AssociatedFile (if
+//      any) whose locale matches the `display_names_locale` field of the
+//      `ImageClassifierOptions` used at creation time ("en" by default, i.e.
+//      English). If none of these are available, only the `index` field of the
+//      results will be filled.
+//
+// An example of such model can be found at:
+// https://tfhub.dev/bohemian-visual-recognition-alliance/lite-model/models/mushroom-identification_v1/1
+//
+// A CLI demo tool is available for easily trying out this API, and provides
+// example usage. See:
+// examples/task/vision/desktop/image_classifier_demo.cc
+class ImageClassifier : public BaseVisionTaskApi<ClassificationResult> {
+ public:
+  using BaseVisionTaskApi::BaseVisionTaskApi;
+
+  // Creates an ImageClassifier from the provided options. A non-default
+  // OpResolver can be specified in order to support custom Ops or specify a
+  // subset of built-in Ops.
+  static tflite::support::StatusOr<std::unique_ptr<ImageClassifier>>
+  CreateFromOptions(
+      const ImageClassifierOptions& options,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Performs actual classification on the provided FrameBuffer.
+  //
+  // The FrameBuffer can be of any size and any of the supported formats, i.e.
+  // RGBA, RGB, NV12, NV21, YV12, YV21. It is automatically pre-processed before
+  // inference in order to (and in this order):
+  // - resize it (with bilinear interpolation, aspect-ratio *not* preserved) to
+  //   the dimensions of the model input tensor,
+  // - convert it to the colorspace of the input tensor (i.e. RGB, which is the
+  //   only supported colorspace for now),
+  // - rotate it according to its `Orientation` so that inference is performed
+  //   on an "upright" image.
+  tflite::support::StatusOr<ClassificationResult> Classify(
+      const FrameBuffer& frame_buffer);
+
+  // Same as above, except that the classification is performed based on the
+  // input region of interest. Cropping according to this region of interest is
+  // prepended to the pre-processing operations.
+  //
+  // IMPORTANT: as a consequence of cropping occurring first, the provided
+  // region of interest is expressed in the unrotated frame of reference
+  // coordinates system, i.e. in `[0, frame_buffer.width) x [0,
+  // frame_buffer.height)`, which are the dimensions of the underlying
+  // `frame_buffer` data before any `Orientation` flag gets applied. Also, the
+  // region of interest is not clamped, so this method will return a non-ok
+  // status if the region is out of these bounds.
+  tflite::support::StatusOr<ClassificationResult> Classify(
+      const FrameBuffer& frame_buffer, const BoundingBox& roi);
+
+ protected:
+  // The options used to build this ImageClassifier.
+  std::unique_ptr<ImageClassifierOptions> options_;
+
+  // The list of classification heads associated with the corresponding output
+  // tensors. Built from TFLite Model Metadata.
+  std::vector<ClassificationHead> classification_heads_;
+
+  // Post-processing to transform the raw model outputs into classification
+  // results.
+  tflite::support::StatusOr<ClassificationResult> Postprocess(
+      const std::vector<const TfLiteTensor*>& output_tensors,
+      const FrameBuffer& frame_buffer, const BoundingBox& roi) override;
+
+  // Performs sanity checks on the provided ImageClassifierOptions.
+  static absl::Status SanityCheckOptions(const ImageClassifierOptions& options);
+
+  // Initializes the ImageClassifier from the provided ImageClassifierOptions,
+  // whose ownership is transferred to this object.
+  absl::Status Init(std::unique_ptr<ImageClassifierOptions> options);
+
+  // Performs pre-initialization actions.
+  virtual absl::Status PreInit();
+  // Performs post-initialization actions.
+  virtual absl::Status PostInit();
+
+ private:
+  // Performs sanity checks on the model outputs and extracts their metadata.
+  absl::Status CheckAndSetOutputs();
+
+  // Performs sanity checks on the class whitelist/blacklist and forms the class
+  // name set.
+  absl::Status CheckAndSetClassNameSet();
+
+  // Initializes the score calibration parameters based on corresponding TFLite
+  // Model Metadata, if any.
+  absl::Status InitScoreCalibrations();
+
+  // Given a ClassificationResult object containing class indices, fills the
+  // name and display name from the label map(s).
+  absl::Status FillResultsFromLabelMaps(ClassificationResult* result);
+
+  // The number of output tensors. This corresponds to the number of
+  // classification heads.
+  int num_outputs_;
+  // Whether the model features quantized inference type (QUANTIZED_UINT8). This
+  // is currently detected by checking if all output tensors data type is uint8.
+  bool has_uint8_outputs_;
+
+  // Set of whitelisted or blacklisted class names.
+  struct ClassNameSet {
+    absl::flat_hash_set<std::string> values;
+    bool is_whitelist;
+  };
+
+  // Whitelisted or blacklisted class names based on provided options at
+  // construction time. These are used to filter out results during
+  // post-processing.
+  ClassNameSet class_name_set_;
+
+  // List of score calibration parameters, if any. Built from TFLite Model
+  // Metadata.
+  std::vector<std::unique_ptr<ScoreCalibration>> score_calibrations_;
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_CLASSIFIER_H_
diff --git a/tensorflow_lite_support/cc/task/vision/image_segmenter.cc b/tensorflow_lite_support/cc/task/vision/image_segmenter.cc
new file mode 100644
index 0000000..4523b66
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/image_segmenter.cc
@@ -0,0 +1,427 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/vision/image_segmenter.h"
+
+#include <algorithm>
+
+#include "absl/memory/memory.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/string_view.h"
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/core/task_api_factory.h"
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+namespace {
+
+using ::absl::StatusCode;
+using ::tflite::TensorMetadata;
+using ::tflite::metadata::ModelMetadataExtractor;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+using ::tflite::task::core::AssertAndReturnTypedTensor;
+using ::tflite::task::core::TaskAPIFactory;
+using ::tflite::task::core::TfLiteEngine;
+
+// The maximum number of labels allowed in the labelmap. This is because so far
+// segmentation masks are stored with 8 bit per pixel (flattened byte array).
+constexpr uint32 kMaxNumClasses = 256;
+
+// TODO(b/)
+// The colormap used to fill `ColoredLabel`-s, as a flattened array of 256 {R,
+// G, B} components.
+constexpr uint8 kColorMap[768] = {
+    0,   0,   0,   128, 0,   0,   0,   128, 0,   128, 128, 0,   0,   0,   128,
+    128, 0,   128, 0,   128, 128, 128, 128, 128, 64,  0,   0,   192, 0,   0,
+    64,  128, 0,   192, 128, 0,   64,  0,   128, 192, 0,   128, 64,  128, 128,
+    192, 128, 128, 0,   64,  0,   128, 64,  0,   0,   192, 0,   128, 192, 0,
+    0,   64,  128, 128, 64,  128, 0,   192, 128, 128, 192, 128, 64,  64,  0,
+    192, 64,  0,   64,  192, 0,   192, 192, 0,   64,  64,  128, 192, 64,  128,
+    64,  192, 128, 192, 192, 128, 0,   0,   64,  128, 0,   64,  0,   128, 64,
+    128, 128, 64,  0,   0,   192, 128, 0,   192, 0,   128, 192, 128, 128, 192,
+    64,  0,   64,  192, 0,   64,  64,  128, 64,  192, 128, 64,  64,  0,   192,
+    192, 0,   192, 64,  128, 192, 192, 128, 192, 0,   64,  64,  128, 64,  64,
+    0,   192, 64,  128, 192, 64,  0,   64,  192, 128, 64,  192, 0,   192, 192,
+    128, 192, 192, 64,  64,  64,  192, 64,  64,  64,  192, 64,  192, 192, 64,
+    64,  64,  192, 192, 64,  192, 64,  192, 192, 192, 192, 192, 32,  0,   0,
+    160, 0,   0,   32,  128, 0,   160, 128, 0,   32,  0,   128, 160, 0,   128,
+    32,  128, 128, 160, 128, 128, 96,  0,   0,   224, 0,   0,   96,  128, 0,
+    224, 128, 0,   96,  0,   128, 224, 0,   128, 96,  128, 128, 224, 128, 128,
+    32,  64,  0,   160, 64,  0,   32,  192, 0,   160, 192, 0,   32,  64,  128,
+    160, 64,  128, 32,  192, 128, 160, 192, 128, 96,  64,  0,   224, 64,  0,
+    96,  192, 0,   224, 192, 0,   96,  64,  128, 224, 64,  128, 96,  192, 128,
+    224, 192, 128, 32,  0,   64,  160, 0,   64,  32,  128, 64,  160, 128, 64,
+    32,  0,   192, 160, 0,   192, 32,  128, 192, 160, 128, 192, 96,  0,   64,
+    224, 0,   64,  96,  128, 64,  224, 128, 64,  96,  0,   192, 224, 0,   192,
+    96,  128, 192, 224, 128, 192, 32,  64,  64,  160, 64,  64,  32,  192, 64,
+    160, 192, 64,  32,  64,  192, 160, 64,  192, 32,  192, 192, 160, 192, 192,
+    96,  64,  64,  224, 64,  64,  96,  192, 64,  224, 192, 64,  96,  64,  192,
+    224, 64,  192, 96,  192, 192, 224, 192, 192, 0,   32,  0,   128, 32,  0,
+    0,   160, 0,   128, 160, 0,   0,   32,  128, 128, 32,  128, 0,   160, 128,
+    128, 160, 128, 64,  32,  0,   192, 32,  0,   64,  160, 0,   192, 160, 0,
+    64,  32,  128, 192, 32,  128, 64,  160, 128, 192, 160, 128, 0,   96,  0,
+    128, 96,  0,   0,   224, 0,   128, 224, 0,   0,   96,  128, 128, 96,  128,
+    0,   224, 128, 128, 224, 128, 64,  96,  0,   192, 96,  0,   64,  224, 0,
+    192, 224, 0,   64,  96,  128, 192, 96,  128, 64,  224, 128, 192, 224, 128,
+    0,   32,  64,  128, 32,  64,  0,   160, 64,  128, 160, 64,  0,   32,  192,
+    128, 32,  192, 0,   160, 192, 128, 160, 192, 64,  32,  64,  192, 32,  64,
+    64,  160, 64,  192, 160, 64,  64,  32,  192, 192, 32,  192, 64,  160, 192,
+    192, 160, 192, 0,   96,  64,  128, 96,  64,  0,   224, 64,  128, 224, 64,
+    0,   96,  192, 128, 96,  192, 0,   224, 192, 128, 224, 192, 64,  96,  64,
+    192, 96,  64,  64,  224, 64,  192, 224, 64,  64,  96,  192, 192, 96,  192,
+    64,  224, 192, 192, 224, 192, 32,  32,  0,   160, 32,  0,   32,  160, 0,
+    160, 160, 0,   32,  32,  128, 160, 32,  128, 32,  160, 128, 160, 160, 128,
+    96,  32,  0,   224, 32,  0,   96,  160, 0,   224, 160, 0,   96,  32,  128,
+    224, 32,  128, 96,  160, 128, 224, 160, 128, 32,  96,  0,   160, 96,  0,
+    32,  224, 0,   160, 224, 0,   32,  96,  128, 160, 96,  128, 32,  224, 128,
+    160, 224, 128, 96,  96,  0,   224, 96,  0,   96,  224, 0,   224, 224, 0,
+    96,  96,  128, 224, 96,  128, 96,  224, 128, 224, 224, 128, 32,  32,  64,
+    160, 32,  64,  32,  160, 64,  160, 160, 64,  32,  32,  192, 160, 32,  192,
+    32,  160, 192, 160, 160, 192, 96,  32,  64,  224, 32,  64,  96,  160, 64,
+    224, 160, 64,  96,  32,  192, 224, 32,  192, 96,  160, 192, 224, 160, 192,
+    32,  96,  64,  160, 96,  64,  32,  224, 64,  160, 224, 64,  32,  96,  192,
+    160, 96,  192, 32,  224, 192, 160, 224, 192, 96,  96,  64,  224, 96,  64,
+    96,  224, 64,  224, 224, 64,  96,  96,  192, 224, 96,  192, 96,  224, 192,
+    224, 224, 192};
+
+StatusOr<std::vector<LabelMapItem>> GetLabelMapIfAny(
+    const ModelMetadataExtractor& metadata_extractor,
+    const TensorMetadata& tensor_metadata, absl::string_view locale) {
+  const std::string labels_filename =
+      ModelMetadataExtractor::FindFirstAssociatedFileName(
+          tensor_metadata, tflite::AssociatedFileType_TENSOR_AXIS_LABELS);
+  if (labels_filename.empty()) {
+    return std::vector<LabelMapItem>();
+  }
+  ASSIGN_OR_RETURN(absl::string_view labels_file,
+                   metadata_extractor.GetAssociatedFile(labels_filename));
+  const std::string display_names_filename =
+      ModelMetadataExtractor::FindFirstAssociatedFileName(
+          tensor_metadata, tflite::AssociatedFileType_TENSOR_AXIS_LABELS,
+          locale);
+  absl::string_view display_names_file = nullptr;
+  if (!display_names_filename.empty()) {
+    ASSIGN_OR_RETURN(display_names_file, metadata_extractor.GetAssociatedFile(
+                                             display_names_filename));
+  }
+  return BuildLabelMapFromFiles(labels_file, display_names_file);
+}
+
+}  // namespace
+
+/* static */
+absl::Status ImageSegmenter::SanityCheckOptions(
+    const ImageSegmenterOptions& options) {
+  if (!options.has_model_file_with_metadata()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Missing mandatory `model_file_with_metadata` field",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.output_type() == ImageSegmenterOptions::UNSPECIFIED) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "ImageSegmenterOptions: `output_type` must not be UNSPECIFIED",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.num_threads() == 0 || options.num_threads() < -1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "`num_threads` must be greater than 0 or equal to -1.",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  return absl::OkStatus();
+}
+
+StatusOr<std::unique_ptr<ImageSegmenter>> ImageSegmenter::CreateFromOptions(
+    const ImageSegmenterOptions& options,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  RETURN_IF_ERROR(SanityCheckOptions(options));
+
+  // Copy options to ensure the ExternalFile outlives the constructed object.
+  auto options_copy = absl::make_unique<ImageSegmenterOptions>(options);
+
+  ASSIGN_OR_RETURN(auto image_segmenter,
+                   TaskAPIFactory::CreateFromExternalFileProto<ImageSegmenter>(
+                       &options_copy->model_file_with_metadata(),
+                       std::move(resolver), options_copy->num_threads()));
+
+  RETURN_IF_ERROR(image_segmenter->Init(std::move(options_copy)));
+
+  return image_segmenter;
+}
+
+absl::Status ImageSegmenter::Init(
+    std::unique_ptr<ImageSegmenterOptions> options) {
+  // Set options.
+  options_ = std::move(options);
+
+  // Perform pre-initialization actions (by default, sets the process engine for
+  // image pre-processing to kLibyuv as a sane default).
+  RETURN_IF_ERROR(PreInit());
+
+  // Sanity check and set inputs and outputs.
+  RETURN_IF_ERROR(CheckAndSetInputs());
+  RETURN_IF_ERROR(CheckAndSetOutputs());
+
+  // Initialize colored_labels_ once and for all.
+  RETURN_IF_ERROR(InitColoredLabels());
+
+  return absl::OkStatus();
+}
+
+absl::Status ImageSegmenter::PreInit() {
+  SetProcessEngine(FrameBufferUtils::ProcessEngine::kLibyuv);
+  return absl::OkStatus();
+}
+
+absl::Status ImageSegmenter::CheckAndSetOutputs() {
+  // First, sanity checks on the model itself.
+  const TfLiteEngine::Interpreter* interpreter = engine_->interpreter();
+
+  // Check the number of output tensors.
+  if (TfLiteEngine::OutputCount(interpreter) != 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Image segmentation models are expected to have only 1 "
+                        "output, found %d",
+                        TfLiteEngine::OutputCount(interpreter)),
+        TfLiteSupportStatus::kInvalidNumOutputTensorsError);
+  }
+  const TfLiteTensor* output_tensor = TfLiteEngine::GetOutput(interpreter, 0);
+
+  // Check tensor dimensions.
+  if (output_tensor->dims->size != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Output tensor is expected to have 4 dimensions, found %d.",
+            output_tensor->dims->size),
+        TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+  }
+  if (output_tensor->dims->data[0] != 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Expected batch size of 1, found %d.",
+                        output_tensor->dims->data[0]),
+        TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+  }
+  output_height_ = output_tensor->dims->data[1];
+  output_width_ = output_tensor->dims->data[2];
+  output_depth_ = output_tensor->dims->data[3];
+  if (output_depth_ > kMaxNumClasses) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Expected at most %d output classes, found %d",
+                        kMaxNumClasses, output_depth_),
+        TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+  }
+
+  // Check tensor type.
+  if (output_tensor->type != kTfLiteFloat32 &&
+      output_tensor->type != kTfLiteUInt8) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Type mismatch for output tensor. Requested one of "
+                        "these types: kTfLiteUint8/kTfLiteFloat32, got %s.",
+                        TfLiteTypeGetName(output_tensor->type)),
+        TfLiteSupportStatus::kInvalidOutputTensorTypeError);
+  }
+  has_uint8_outputs_ = (output_tensor->type == kTfLiteUInt8);
+
+  // Build label map from metadata, if available.
+  const ModelMetadataExtractor* metadata_extractor =
+      engine_->metadata_extractor();
+  const flatbuffers::Vector<flatbuffers::Offset<TensorMetadata>>*
+      output_tensor_metadata = metadata_extractor->GetOutputTensorMetadata();
+  if (output_tensor_metadata != nullptr) {
+    // Check metadata consistency.
+    if (output_tensor_metadata->size() != 1) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Mismatch between number of output tensors (1) and "
+                          "output tensors metadata (%d).",
+                          output_tensor_metadata->size()),
+          TfLiteSupportStatus::kMetadataInconsistencyError);
+    }
+    ASSIGN_OR_RETURN(
+        label_map_,
+        GetLabelMapIfAny(*metadata_extractor, *output_tensor_metadata->Get(0),
+                         options_->display_names_locale()));
+  }
+
+  // If label map is still empty, build a default one.
+  if (label_map_.empty()) {
+    for (int class_index = 0; class_index < output_depth_; ++class_index) {
+      label_map_.emplace_back(LabelMapItem{});
+    }
+  }
+
+  return absl::OkStatus();
+}
+
+absl::Status ImageSegmenter::InitColoredLabels() {
+  for (int i = 0; i < label_map_.size(); ++i) {
+    Segmentation::ColoredLabel colored_label;
+    colored_label.set_r(kColorMap[3 * i]);
+    colored_label.set_g(kColorMap[3 * i + 1]);
+    colored_label.set_b(kColorMap[3 * i + 2]);
+    const LabelMapItem& item = label_map_[i];
+    if (!item.name.empty()) {
+      colored_label.set_class_name(item.name);
+    }
+    if (!item.display_name.empty()) {
+      colored_label.set_display_name(item.display_name);
+    }
+    colored_labels_.push_back(colored_label);
+  }
+  return absl::OkStatus();
+}
+
+StatusOr<SegmentationResult> ImageSegmenter::Segment(
+    const FrameBuffer& frame_buffer) {
+  BoundingBox roi;
+  roi.set_width(frame_buffer.dimension().width);
+  roi.set_height(frame_buffer.dimension().height);
+  return InferWithFallback(frame_buffer, roi);
+}
+
+StatusOr<SegmentationResult> ImageSegmenter::Postprocess(
+    const std::vector<const TfLiteTensor*>& output_tensors,
+    const FrameBuffer& frame_buffer, const BoundingBox& /*roi*/) {
+  if (output_tensors.size() != 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Expected 1 output tensors, found %d",
+                        output_tensors.size()));
+  }
+  const TfLiteTensor* output_tensor = output_tensors[0];
+
+  SegmentationResult result;
+  Segmentation* segmentation = result.add_segmentation();
+  *segmentation->mutable_colored_labels() = {colored_labels_.begin(),
+                                             colored_labels_.end()};
+
+  // The output tensor has orientation `frame_buffer.orientation()`, as it has
+  // been produced from the pre-processed frame.
+  FrameBuffer::Orientation tensor_orientation = frame_buffer.orientation();
+  // The output tensor always has size `output_width_ x output_height_`
+  FrameBuffer::Dimension tensor_dimension = {output_width_, output_height_};
+
+  // The masks to produce from the output tensor need to be re-oriented in the
+  // unrotated frame of reference coordinates system, i.e. kTopLeft.
+  FrameBuffer::Orientation mask_orientation =
+      FrameBuffer::Orientation::kTopLeft;
+  // They may thus have swapped dimensions compared to the tensor if the
+  // rotation is 90° or 270°.
+  FrameBuffer::Dimension mask_dimension(tensor_dimension);
+  if (RequireDimensionSwap(frame_buffer.orientation(),
+                           FrameBuffer::Orientation::kTopLeft)) {
+    mask_dimension.Swap();
+  }
+  segmentation->set_width(mask_dimension.width);
+  segmentation->set_height(mask_dimension.height);
+
+  // XY coordinates in the tensor, to be computed from mask_x and mask_y below.
+  int tensor_x;
+  int tensor_y;
+
+  if (options_->output_type() == ImageSegmenterOptions::CATEGORY_MASK) {
+    auto* category_mask = segmentation->mutable_category_mask();
+    category_mask->resize(mask_dimension.width * mask_dimension.height);
+    int pixel_offset = 0;
+    for (int mask_y = 0; mask_y < mask_dimension.height; ++mask_y) {
+      for (int mask_x = 0; mask_x < mask_dimension.width; ++mask_x) {
+        // Compute the coordinates (tensor_x, tensor_y) in the tensor with
+        // tensor_orientation = frame_buffer.orientation() corresponding to the
+        // coordinates (mask_x, mask_y) in the mask being filled with
+        // mask_orientation = kTopLeft, i.e. the orientation of the unrotated
+        // frame of reference.
+        OrientCoordinates(/*from_x=*/mask_x,
+                          /*from_y=*/mask_y,
+                          /*from_orientation=*/mask_orientation,
+                          /*to_orientation=*/tensor_orientation,
+                          /*from_dimension=*/mask_dimension,
+                          /*to_x=*/&tensor_x,
+                          /*to_y=*/&tensor_y);
+        int class_index = 0;
+        float max_confidence = 0.0f;
+        for (int d = 0; d < output_depth_; ++d) {
+          const float confidence =
+              GetOutputConfidence(*output_tensor, tensor_x, tensor_y, d);
+          if (confidence > max_confidence) {
+            class_index = d;
+            max_confidence = confidence;
+          }
+        }
+        (*category_mask)[pixel_offset++] = static_cast<char>(class_index);
+      }
+    }
+  } else if (options_->output_type() ==
+             ImageSegmenterOptions::CONFIDENCE_MASK) {
+    auto* confidence_masks = segmentation->mutable_confidence_masks();
+    for (int d = 0; d < output_depth_; ++d) {
+      confidence_masks->add_confidence_mask();
+    }
+    for (int mask_y = 0; mask_y < segmentation->height(); ++mask_y) {
+      for (int mask_x = 0; mask_x < segmentation->width(); ++mask_x) {
+        // See above.
+        OrientCoordinates(/*from_x=*/mask_x,
+                          /*from_y=*/mask_y,
+                          /*from_orientation=*/mask_orientation,
+                          /*to_orientation=*/tensor_orientation,
+                          /*from_dimension=*/mask_dimension,
+                          /*to_x=*/&tensor_x,
+                          /*to_y=*/&tensor_y);
+        for (int d = 0; d < output_depth_; ++d) {
+          confidence_masks->mutable_confidence_mask(d)->add_value(
+              GetOutputConfidence(*output_tensor, tensor_x, tensor_y, d));
+        }
+      }
+    }
+  }
+
+  return result;
+}
+
+float ImageSegmenter::GetOutputConfidence(const TfLiteTensor& output_tensor,
+                                          int x, int y, int depth) {
+  int index = output_width_ * output_depth_ * y + output_depth_ * x + depth;
+  if (has_uint8_outputs_) {
+    const uint8* data = AssertAndReturnTypedTensor<uint8>(&output_tensor);
+    return output_tensor.params.scale *
+           (static_cast<int>(data[index]) - output_tensor.params.zero_point);
+  } else {
+    const float* data = AssertAndReturnTypedTensor<float>(&output_tensor);
+    return data[index];
+  }
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/image_segmenter.h b/tensorflow_lite_support/cc/task/vision/image_segmenter.h
new file mode 100644
index 0000000..663ddb7
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/image_segmenter.h
@@ -0,0 +1,172 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_SEGMENTER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_SEGMENTER_H_
+
+#include <memory>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+#include "tensorflow_lite_support/cc/task/vision/core/base_vision_task_api.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/core/label_map_item.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/segmentations_proto_inc.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Performs segmentation on images.
+//
+// The API expects a TFLite model with optional, but strongly recommended,
+// TFLite Model Metadata.
+//
+// Input tensor:
+//   (kTfLiteUInt8/kTfLiteFloat32)
+//    - image input of size `[batch x height x width x channels]`.
+//    - batch inference is not supported (`batch` is required to be 1).
+//    - only RGB inputs are supported (`channels` is required to be 3).
+//    - if type is kTfLiteFloat32, NormalizationOptions are required to be
+//      attached to the metadata for input normalization.
+// Output tensor:
+//   (kTfLiteUInt8/kTfLiteFloat32)
+//    - tensor of size `[batch x mask_height x mask_width x num_classes]`, where
+//      `batch` is required to be 1, `mask_width` and `mask_height` are the
+//      dimensions of the segmentation masks produced by the model, and
+//      `num_classes` is the number of classes supported by the model.
+//    - optional (but recommended) label map(s) can be attached as
+//      AssociatedFile-s with type TENSOR_AXIS_LABELS, containing one label per
+//      line. The first such AssociatedFile (if any) is used to fill the
+//      `class_name` field of the results. The `display_name` field is filled
+//      from the AssociatedFile (if any) whose locale matches the
+//      `display_names_locale` field of the `ImageSegmenterOptions` used at
+//      creation time ("en" by default, i.e. English). If none of these are
+//      available, only the `index` field of the results will be filled.
+//
+// An example of such model can be found at:
+// https://tfhub.dev/tensorflow/lite-model/deeplabv3/1/metadata/1
+//
+// A CLI demo tool is available for easily trying out this API, and provides
+// example usage. See:
+// examples/task/vision/desktop/image_segmenter_demo.cc
+class ImageSegmenter : public BaseVisionTaskApi<SegmentationResult> {
+ public:
+  using BaseVisionTaskApi::BaseVisionTaskApi;
+
+  // Creates an ImageSegmenter from the provided options. A non-default
+  // OpResolver can be specified in order to support custom Ops or specify a
+  // subset of built-in Ops.
+  static tflite::support::StatusOr<std::unique_ptr<ImageSegmenter>>
+  CreateFromOptions(
+      const ImageSegmenterOptions& options,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Performs actual segmentation on the provided FrameBuffer.
+  //
+  // The FrameBuffer can be of any size and any of the supported formats, i.e.
+  // RGBA, RGB, NV12, NV21, YV12, YV21. It is automatically pre-processed before
+  // inference in order to (and in this order):
+  // - resize it (with bilinear interpolation, aspect-ratio *not* preserved) to
+  //   the dimensions of the model input tensor,
+  // - convert it to the colorspace of the input tensor (i.e. RGB, which is the
+  //   only supported colorspace for now),
+  // - rotate it according to its `Orientation` so that inference is performed
+  //   on an "upright" image.
+  //
+  // IMPORTANT: the returned segmentation masks are not direcly suited for
+  // display, in particular:
+  // * they are relative to the unrotated input frame, i.e. *not* taking into
+  //   account the `Orientation` flag of the input FrameBuffer,
+  // * their dimensions are intrinsic to the model, i.e. *not* dependent on the
+  //   input FrameBuffer dimensions.
+  //
+  // Example of such post-processing, assuming:
+  // * an input FrameBuffer with width=640, height=480, orientation=kLeftBottom
+  //   (i.e. the image will be rotated 90° clockwise during preprocessing to
+  //   make it "upright"),
+  // * a model outputting masks of size 224x224.
+  // In order to be directly displayable on top of the input image assumed to
+  // be displayed *with* the `Orientation` flag taken into account according to
+  // the EXIF specification (http://jpegclub.org/exif_orientation.html), the
+  // masks need to be:
+  // * re-scaled to 640 x 480,
+  // * then rotated 90° clockwise.
+  tflite::support::StatusOr<SegmentationResult> Segment(
+      const FrameBuffer& frame_buffer);
+
+ protected:
+  // Post-processing to transform the raw model outputs into segmentation
+  // results.
+  tflite::support::StatusOr<SegmentationResult> Postprocess(
+      const std::vector<const TfLiteTensor*>& output_tensors,
+      const FrameBuffer& frame_buffer, const BoundingBox& roi) override;
+
+  // Performs sanity checks on the provided ImageSegmenterOptions.
+  static absl::Status SanityCheckOptions(const ImageSegmenterOptions& options);
+
+  // Initializes the Segmenter from the provided ImageSegmenterOptions, whose
+  // ownership is transferred to this object.
+  absl::Status Init(std::unique_ptr<ImageSegmenterOptions> options);
+
+  // Performs pre-initialization actions.
+  virtual absl::Status PreInit();
+
+  // The options used for building this image segmenter.
+  std::unique_ptr<ImageSegmenterOptions> options_;
+
+  // The label map, extracted from the TFLite Model Metadata.
+  std::vector<LabelMapItem> label_map_;
+
+ private:
+  // Performs sanity checks on the model outputs and extracts their metadata.
+  absl::Status CheckAndSetOutputs();
+
+  // Initializes the colored labels list from `label_map_` and stores it in
+  // `colored_labels_`.
+  absl::Status InitColoredLabels();
+
+  // Returns the output confidence at coordinates {x, y, depth}, dequantizing
+  // on-the-fly if needed (i.e. if `has_uint8_outputs_` is true).
+  float GetOutputConfidence(const TfLiteTensor& output_tensor, int x, int y,
+                            int depth);
+
+  // Prebuilt list of ColoredLabel attached to each Segmentation result. The
+  // i-th item in this list corresponds to the i-th label map item.
+  std::vector<Segmentation::ColoredLabel> colored_labels_;
+
+  // Whether the model features quantized inference type (QUANTIZED_UINT8). This
+  // is currently detected by checking if all output tensors data type is uint8.
+  bool has_uint8_outputs_;
+
+  // Expected output width.
+  int output_width_;
+  // Expected output height.
+  int output_height_;
+  // Expected output depth. This corresponds to the number of supported classes.
+  int output_depth_;
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_SEGMENTER_H_
diff --git a/tensorflow_lite_support/cc/task/vision/object_detector.cc b/tensorflow_lite_support/cc/task/vision/object_detector.cc
new file mode 100644
index 0000000..22ec301
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/object_detector.cc
@@ -0,0 +1,549 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/vision/object_detector.h"
+
+#include <algorithm>
+#include <limits>
+#include <vector>
+
+#include "absl/memory/memory.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/interpreter.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/core/task_api_factory.h"
+#include "tensorflow_lite_support/cc/task/core/task_utils.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+#include "tensorflow_lite_support/cc/task/vision/core/label_map_item.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+namespace {
+
+using ::absl::StatusCode;
+using ::tflite::BoundingBoxProperties;
+using ::tflite::ContentProperties;
+using ::tflite::ContentProperties_BoundingBoxProperties;
+using ::tflite::EnumNameContentProperties;
+using ::tflite::ProcessUnit;
+using ::tflite::ProcessUnitOptions_ScoreThresholdingOptions;
+using ::tflite::TensorMetadata;
+using ::tflite::metadata::ModelMetadataExtractor;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+using ::tflite::task::core::AssertAndReturnTypedTensor;
+using ::tflite::task::core::TaskAPIFactory;
+using ::tflite::task::core::TfLiteEngine;
+
+// The expected number of dimensions of the 4 output tensors, representing in
+// that order: locations, classes, scores, num_results.
+static constexpr int kOutputTensorsExpectedDims[4] = {3, 2, 2, 1};
+
+StatusOr<const BoundingBoxProperties*> GetBoundingBoxProperties(
+    const TensorMetadata& tensor_metadata) {
+  if (tensor_metadata.content() == nullptr ||
+      tensor_metadata.content()->content_properties() == nullptr) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Expected BoundingBoxProperties for tensor %s, found none.",
+            tensor_metadata.name() ? tensor_metadata.name()->str() : "#0"),
+        TfLiteSupportStatus::kMetadataInvalidContentPropertiesError);
+  }
+
+  ContentProperties type = tensor_metadata.content()->content_properties_type();
+  if (type != ContentProperties_BoundingBoxProperties) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Expected BoundingBoxProperties for tensor %s, found %s.",
+            tensor_metadata.name() ? tensor_metadata.name()->str() : "#0",
+            EnumNameContentProperties(type)),
+        TfLiteSupportStatus::kMetadataInvalidContentPropertiesError);
+  }
+
+  const BoundingBoxProperties* properties =
+      tensor_metadata.content()->content_properties_as_BoundingBoxProperties();
+
+  // Mobile SSD only supports "BOUNDARIES" bounding box type.
+  if (properties->type() != tflite::BoundingBoxType_BOUNDARIES) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Mobile SSD only supports BoundingBoxType BOUNDARIES, found %s",
+            tflite::EnumNameBoundingBoxType(properties->type())),
+        TfLiteSupportStatus::kMetadataInvalidContentPropertiesError);
+  }
+
+  // Mobile SSD only supports "RATIO" coordinates type.
+  if (properties->coordinate_type() != tflite::CoordinateType_RATIO) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Mobile SSD only supports CoordinateType RATIO, found %s",
+            tflite::EnumNameCoordinateType(properties->coordinate_type())),
+        TfLiteSupportStatus::kMetadataInvalidContentPropertiesError);
+  }
+
+  // Index is optional, but must contain 4 values if present.
+  if (properties->index() != nullptr && properties->index()->size() != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Expected BoundingBoxProperties index to contain 4 values, found "
+            "%d",
+            properties->index()->size()),
+        TfLiteSupportStatus::kMetadataInvalidContentPropertiesError);
+  }
+
+  return properties;
+}
+
+StatusOr<std::vector<LabelMapItem>> GetLabelMapIfAny(
+    const ModelMetadataExtractor& metadata_extractor,
+    const TensorMetadata& tensor_metadata, absl::string_view locale) {
+  const std::string labels_filename =
+      ModelMetadataExtractor::FindFirstAssociatedFileName(
+          tensor_metadata, tflite::AssociatedFileType_TENSOR_VALUE_LABELS);
+  if (labels_filename.empty()) {
+    return std::vector<LabelMapItem>();
+  }
+  ASSIGN_OR_RETURN(absl::string_view labels_file,
+                   metadata_extractor.GetAssociatedFile(labels_filename));
+  const std::string display_names_filename =
+      ModelMetadataExtractor::FindFirstAssociatedFileName(
+          tensor_metadata, tflite::AssociatedFileType_TENSOR_VALUE_LABELS,
+          locale);
+  absl::string_view display_names_file = nullptr;
+  if (!display_names_filename.empty()) {
+    ASSIGN_OR_RETURN(display_names_file, metadata_extractor.GetAssociatedFile(
+                                             display_names_filename));
+  }
+  return BuildLabelMapFromFiles(labels_file, display_names_file);
+}
+
+StatusOr<float> GetScoreThreshold(
+    const ModelMetadataExtractor& metadata_extractor,
+    const TensorMetadata& tensor_metadata) {
+  ASSIGN_OR_RETURN(
+      const ProcessUnit* score_thresholding_process_unit,
+      metadata_extractor.FindFirstProcessUnit(
+          tensor_metadata, ProcessUnitOptions_ScoreThresholdingOptions));
+  if (score_thresholding_process_unit == nullptr) {
+    return std::numeric_limits<float>::lowest();
+  }
+  return score_thresholding_process_unit->options_as_ScoreThresholdingOptions()
+      ->global_score_threshold();
+}
+
+absl::Status SanityCheckOutputTensors(
+    const std::vector<const TfLiteTensor*>& output_tensors) {
+  if (output_tensors.size() != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Expected 4 output tensors, found %d",
+                        output_tensors.size()));
+  }
+
+  // Get number of results.
+  if (output_tensors[3]->dims->data[0] != 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat(
+            "Expected tensor with dimensions [1] at index 3, found [%d]",
+            output_tensors[3]->dims->data[0]));
+  }
+  int num_results =
+      static_cast<int>(AssertAndReturnTypedTensor<float>(output_tensors[3])[0]);
+
+  // Check dimensions for the other tensors are correct.
+  if (output_tensors[0]->dims->data[0] != 1 ||
+      output_tensors[0]->dims->data[1] != num_results ||
+      output_tensors[0]->dims->data[2] != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat(
+            "Expected locations tensor with dimensions [1,%d,4] at index 0, "
+            "found [%d,%d,%d].",
+            num_results, output_tensors[0]->dims->data[0],
+            output_tensors[0]->dims->data[1],
+            output_tensors[0]->dims->data[2]));
+  }
+  if (output_tensors[1]->dims->data[0] != 1 ||
+      output_tensors[1]->dims->data[1] != num_results) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat(
+            "Expected classes tensor with dimensions [1,%d] at index 1, "
+            "found [%d,%d].",
+            num_results, output_tensors[1]->dims->data[0],
+            output_tensors[1]->dims->data[1]));
+  }
+  if (output_tensors[2]->dims->data[0] != 1 ||
+      output_tensors[2]->dims->data[1] != num_results) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat(
+            "Expected scores tensor with dimensions [1,%d] at index 2, "
+            "found [%d,%d].",
+            num_results, output_tensors[2]->dims->data[0],
+            output_tensors[2]->dims->data[1]));
+  }
+
+  return absl::OkStatus();
+}
+
+}  // namespace
+
+/* static */
+absl::Status ObjectDetector::SanityCheckOptions(
+    const ObjectDetectorOptions& options) {
+  if (!options.has_model_file_with_metadata()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Missing mandatory `model_file_with_metadata` field",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.max_results() == 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Invalid `max_results` option: value must be != 0",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.class_name_whitelist_size() > 0 &&
+      options.class_name_blacklist_size() > 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "`class_name_whitelist` and `class_name_blacklist` are mutually "
+        "exclusive options.",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (options.num_threads() == 0 || options.num_threads() < -1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "`num_threads` must be greater than 0 or equal to -1.",
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  return absl::OkStatus();
+}
+
+/* static */
+StatusOr<std::unique_ptr<ObjectDetector>> ObjectDetector::CreateFromOptions(
+    const ObjectDetectorOptions& options,
+    std::unique_ptr<tflite::OpResolver> resolver) {
+  RETURN_IF_ERROR(SanityCheckOptions(options));
+
+  // Copy options to ensure the ExternalFile outlives the constructed object.
+  auto options_copy = absl::make_unique<ObjectDetectorOptions>(options);
+
+  ASSIGN_OR_RETURN(auto object_detector,
+                   TaskAPIFactory::CreateFromExternalFileProto<ObjectDetector>(
+                       &options_copy->model_file_with_metadata(),
+                       std::move(resolver), options_copy->num_threads()));
+
+  RETURN_IF_ERROR(object_detector->Init(std::move(options_copy)));
+
+  return object_detector;
+}
+
+absl::Status ObjectDetector::Init(
+    std::unique_ptr<ObjectDetectorOptions> options) {
+  // Set options.
+  options_ = std::move(options);
+
+  // Perform pre-initialization actions (by default, sets the process engine for
+  // image pre-processing to kLibyuv as a sane default).
+  RETURN_IF_ERROR(PreInit());
+
+  // Sanity check and set inputs and outputs.
+  RETURN_IF_ERROR(CheckAndSetInputs());
+  RETURN_IF_ERROR(CheckAndSetOutputs());
+
+  // Initialize class whitelisting/blacklisting, if any.
+  RETURN_IF_ERROR(CheckAndSetClassIndexSet());
+
+  return absl::OkStatus();
+}
+
+absl::Status ObjectDetector::PreInit() {
+  SetProcessEngine(FrameBufferUtils::ProcessEngine::kLibyuv);
+  return absl::OkStatus();
+}
+
+absl::Status ObjectDetector::CheckAndSetOutputs() {
+  // First, sanity checks on the model itself.
+  const TfLiteEngine::Interpreter* interpreter = engine_->interpreter();
+  // Check the number of output tensors.
+  if (TfLiteEngine::OutputCount(interpreter) != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Mobile SSD models are expected to have exactly 4 "
+                        "outputs, found %d",
+                        TfLiteEngine::OutputCount(interpreter)),
+        TfLiteSupportStatus::kInvalidNumOutputTensorsError);
+  }
+  // Check tensor dimensions and batch size.
+  for (int i = 0; i < 4; ++i) {
+    const TfLiteTensor* tensor = TfLiteEngine::GetOutput(interpreter, i);
+    if (tensor->dims->size != kOutputTensorsExpectedDims[i]) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Output tensor at index %d is expected to "
+                          "have %d dimensions, found %d.",
+                          i, kOutputTensorsExpectedDims[i], tensor->dims->size),
+          TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+    }
+    if (tensor->dims->data[0] != 1) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat("Expected batch size of 1, found %d.",
+                          tensor->dims->data[0]),
+          TfLiteSupportStatus::kInvalidOutputTensorDimensionsError);
+    }
+  }
+
+  // Now, perform sanity checks and extract metadata.
+  const ModelMetadataExtractor* metadata_extractor =
+      engine_->metadata_extractor();
+  // Check that metadata is available.
+  if (metadata_extractor->GetModelMetadata() == nullptr ||
+      metadata_extractor->GetModelMetadata()->subgraph_metadata() == nullptr) {
+    return CreateStatusWithPayload(StatusCode::kInvalidArgument,
+                                   "Object detection models require TFLite "
+                                   "Model Metadata but none was found",
+                                   TfLiteSupportStatus::kMetadataNotFoundError);
+  }
+  // Check output tensor metadata is present and consistent with model.
+  auto output_tensors_metadata = metadata_extractor->GetOutputTensorMetadata();
+  if (output_tensors_metadata == nullptr ||
+      output_tensors_metadata->size() != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Mismatch between number of output tensors (4) and output tensors "
+            "metadata (%d).",
+            output_tensors_metadata == nullptr
+                ? 0
+                : output_tensors_metadata->size()),
+        TfLiteSupportStatus::kMetadataInconsistencyError);
+  }
+
+  // Extract mandatory BoundingBoxProperties for easier access at
+  // post-processing time, performing sanity checks on the fly.
+  ASSIGN_OR_RETURN(const BoundingBoxProperties* bounding_box_properties,
+                   GetBoundingBoxProperties(*output_tensors_metadata->Get(0)));
+  if (bounding_box_properties->index() == nullptr) {
+    bounding_box_corners_order_ = {0, 1, 2, 3};
+  } else {
+    auto bounding_box_index = bounding_box_properties->index();
+    bounding_box_corners_order_ = {
+        bounding_box_index->Get(0),
+        bounding_box_index->Get(1),
+        bounding_box_index->Get(2),
+        bounding_box_index->Get(3),
+    };
+  }
+
+  // Build label map (if available) from metadata.
+  ASSIGN_OR_RETURN(
+      label_map_,
+      GetLabelMapIfAny(*metadata_extractor, *output_tensors_metadata->Get(1),
+                       options_->display_names_locale()));
+
+  // Set score threshold.
+  if (options_->has_score_threshold()) {
+    score_threshold_ = options_->score_threshold();
+  } else {
+    ASSIGN_OR_RETURN(score_threshold_,
+                     GetScoreThreshold(*metadata_extractor,
+                                       *output_tensors_metadata->Get(2)));
+  }
+
+  return absl::OkStatus();
+}
+
+absl::Status ObjectDetector::CheckAndSetClassIndexSet() {
+  // Exit early if no blacklist/whitelist.
+  if (options_->class_name_blacklist_size() == 0 &&
+      options_->class_name_whitelist_size() == 0) {
+    return absl::OkStatus();
+  }
+  // Label map is mandatory.
+  if (label_map_.empty()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Using `class_name_whitelist` or `class_name_blacklist` requires "
+        "labels to be present in the TFLite Model Metadata but none was found.",
+        TfLiteSupportStatus::kMetadataMissingLabelsError);
+  }
+
+  class_index_set_.is_whitelist = options_->class_name_whitelist_size() > 0;
+  const auto& class_names = class_index_set_.is_whitelist
+                                ? options_->class_name_whitelist()
+                                : options_->class_name_blacklist();
+  class_index_set_.values.clear();
+  for (const auto& class_name : class_names) {
+    int index = -1;
+    for (int i = 0; i < label_map_.size(); ++i) {
+      if (label_map_[i].name == class_name) {
+        index = i;
+        break;
+      }
+    }
+    // Ignore duplicate or unknown classes.
+    if (index < 0 || class_index_set_.values.contains(index)) {
+      continue;
+    }
+    class_index_set_.values.insert(index);
+  }
+
+  if (class_index_set_.values.empty()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat(
+            "Invalid class names specified via `class_name_%s`: none match "
+            "with model labels.",
+            class_index_set_.is_whitelist ? "whitelist" : "blacklist"),
+        TfLiteSupportStatus::kInvalidArgumentError);
+  }
+
+  return absl::OkStatus();
+}
+
+StatusOr<DetectionResult> ObjectDetector::Detect(
+    const FrameBuffer& frame_buffer) {
+  BoundingBox roi;
+  roi.set_width(frame_buffer.dimension().width);
+  roi.set_height(frame_buffer.dimension().height);
+  // Rely on `Infer` instead of `InferWithFallback` as DetectionPostprocessing
+  // op doesn't support hardware acceleration at the time.
+  return Infer(frame_buffer, roi);
+}
+
+StatusOr<DetectionResult> ObjectDetector::Postprocess(
+    const std::vector<const TfLiteTensor*>& output_tensors,
+    const FrameBuffer& frame_buffer, const BoundingBox& /*roi*/) {
+  // Most of the checks here should never happen, as outputs have been validated
+  // at construction time. Checking nonetheless and returning internal errors if
+  // something bad happens.
+  RETURN_IF_ERROR(SanityCheckOutputTensors(output_tensors));
+
+  // Get number of available results.
+  const int num_results =
+      static_cast<int>(AssertAndReturnTypedTensor<float>(output_tensors[3])[0]);
+  // Compute number of max results to return.
+  const int max_results = options_->max_results() > 0
+                              ? std::min(options_->max_results(), num_results)
+                              : num_results;
+  // The dimensions of the upright (i.e. rotated according to its orientation)
+  // input frame.
+  FrameBuffer::Dimension upright_input_frame_dimensions =
+      frame_buffer.dimension();
+  if (RequireDimensionSwap(frame_buffer.orientation(),
+                           FrameBuffer::Orientation::kTopLeft)) {
+    upright_input_frame_dimensions.Swap();
+  }
+
+  const float* locations = AssertAndReturnTypedTensor<float>(output_tensors[0]);
+  const float* classes = AssertAndReturnTypedTensor<float>(output_tensors[1]);
+  const float* scores = AssertAndReturnTypedTensor<float>(output_tensors[2]);
+  DetectionResult results;
+  for (int i = 0; i < num_results; ++i) {
+    const int class_index = static_cast<int>(classes[i]);
+    const float score = scores[i];
+    if (!IsClassIndexAllowed(class_index) || score < score_threshold_) {
+      continue;
+    }
+    Detection* detection = results.add_detections();
+    // Denormalize the bounding box cooordinates in the upright frame
+    // coordinates system, then rotate back from frame_buffer.orientation() to
+    // the unrotated frame of reference coordinates system (i.e. with
+    // orientation = kTopLeft).
+    *detection->mutable_bounding_box() = OrientAndDenormalizeBoundingBox(
+        /*from_left=*/locations[4 * i + bounding_box_corners_order_[0]],
+        /*from_top=*/locations[4 * i + bounding_box_corners_order_[1]],
+        /*from_right=*/locations[4 * i + bounding_box_corners_order_[2]],
+        /*from_bottom=*/locations[4 * i + bounding_box_corners_order_[3]],
+        /*from_orientation=*/frame_buffer.orientation(),
+        /*to_orientation=*/FrameBuffer::Orientation::kTopLeft,
+        /*from_dimension=*/upright_input_frame_dimensions);
+    Class* detection_class = detection->add_classes();
+    detection_class->set_index(class_index);
+    detection_class->set_score(score);
+    if (results.detections_size() == max_results) {
+      break;
+    }
+  }
+
+  if (!label_map_.empty()) {
+    RETURN_IF_ERROR(FillResultsFromLabelMap(&results));
+  }
+
+  return results;
+}
+
+bool ObjectDetector::IsClassIndexAllowed(int class_index) {
+  if (class_index_set_.values.empty()) {
+    return true;
+  }
+  if (class_index_set_.is_whitelist) {
+    return class_index_set_.values.contains(class_index);
+  } else {
+    return !class_index_set_.values.contains(class_index);
+  }
+}
+
+absl::Status ObjectDetector::FillResultsFromLabelMap(DetectionResult* result) {
+  for (int i = 0; i < result->detections_size(); ++i) {
+    Detection* detection = result->mutable_detections(i);
+    for (int j = 0; j < detection->classes_size(); ++j) {
+      Class* detection_class = detection->mutable_classes(j);
+      const int index = detection_class->index();
+      if (index >= label_map_.size()) {
+        return CreateStatusWithPayload(
+            StatusCode::kInvalidArgument,
+            absl::StrFormat(
+                "Label map does not contain enough elements: model returned "
+                "class index %d but label map only contains %d elements.",
+                index, label_map_.size()),
+            TfLiteSupportStatus::kMetadataInconsistencyError);
+      }
+      std::string name = label_map_[index].name;
+      if (!name.empty()) {
+        detection_class->set_class_name(name);
+      }
+      std::string display_name = label_map_[index].display_name;
+      if (!display_name.empty()) {
+        detection_class->set_display_name(display_name);
+      }
+    }
+  }
+  return absl::OkStatus();
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/object_detector.h b/tensorflow_lite_support/cc/task/vision/object_detector.h
new file mode 100644
index 0000000..2bd220b
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/object_detector.h
@@ -0,0 +1,186 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_OBJECT_DETECTOR_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_OBJECT_DETECTOR_H_
+
+#include <memory>
+
+#include "absl/container/flat_hash_set.h"
+#include "absl/status/status.h"
+#include "tensorflow/lite/core/api/op_resolver.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+#include "tensorflow_lite_support/cc/task/vision/core/base_vision_task_api.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/core/label_map_item.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/detections_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/object_detector_options_proto_inc.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Performs object detection on images.
+//
+// The API expects a TFLite model with mandatory TFLite Model Metadata.
+//
+// Input tensor:
+//   (kTfLiteUInt8/kTfLiteFloat32)
+//    - image input of size `[batch x height x width x channels]`.
+//    - batch inference is not supported (`batch` is required to be 1).
+//    - only RGB inputs are supported (`channels` is required to be 3).
+//    - if type is kTfLiteFloat32, NormalizationOptions are required to be
+//      attached to the metadata for input normalization.
+// Output tensors must be the 4 outputs of a `DetectionPostProcess` op, i.e:
+//  (kTfLiteFloat32)
+//   - locations tensor of size `[num_results x 4]`, the inner array
+//     representing bounding boxes in the form [top, left, right, bottom].
+//   - BoundingBoxProperties are required to be attached to the metadata
+//     and must specify type=BOUNDARIES and coordinate_type=RATIO.
+//  (kTfLiteFloat32)
+//   - classes tensor of size `[num_results]`, each value representing the
+//     integer index of a class.
+//    - optional (but recommended) label map(s) can be attached as
+//      AssociatedFile-s with type TENSOR_VALUE_LABELS, containing one label per
+//      line. The first such AssociatedFile (if any) is used to fill the
+//      `class_name` field of the results. The `display_name` field is filled
+//      from the AssociatedFile (if any) whose locale matches the
+//      `display_names_locale` field of the `ObjectDetectorOptions` used at
+//      creation time ("en" by default, i.e. English). If none of these are
+//      available, only the `index` field of the results will be filled.
+//  (kTfLiteFloat32)
+//   - scores tensor of size `[num_results]`, each value representing the score
+//     of the detected object.
+//  (kTfLiteFloat32)
+//   - integer num_results as a tensor of size `[1]`
+//
+// An example of such model can be found at:
+// https://tfhub.dev/google/lite-model/object_detection/mobile_object_localizer_v1/1/metadata/1
+//
+// A CLI demo tool is available for easily trying out this API, and provides
+// example usage. See:
+// examples/task/vision/desktop/object_detector_demo.cc
+class ObjectDetector : public BaseVisionTaskApi<DetectionResult> {
+ public:
+  using BaseVisionTaskApi::BaseVisionTaskApi;
+
+  // Creates an ObjectDetector from the provided options. A non-default
+  // OpResolver can be specified in order to support custom Ops or specify a
+  // subset of built-in Ops.
+  static tflite::support::StatusOr<std::unique_ptr<ObjectDetector>>
+  CreateFromOptions(
+      const ObjectDetectorOptions& options,
+      std::unique_ptr<tflite::OpResolver> resolver =
+          absl::make_unique<tflite::ops::builtin::BuiltinOpResolver>());
+
+  // Performs actual detection on the provided FrameBuffer.
+  //
+  // The FrameBuffer can be of any size and any of the supported formats, i.e.
+  // RGBA, RGB, NV12, NV21, YV12, YV21. It is automatically pre-processed
+  // before inference in order to (and in this order):
+  // - resize it (with bilinear interpolation, aspect-ratio *not* preserved) to
+  //   the dimensions of the model input tensor,
+  // - convert it to the colorspace of the input tensor (i.e. RGB, which is the
+  //   only supported colorspace for now),
+  // - rotate it according to its `Orientation` so that inference is performed
+  //   on an "upright" image.
+  //
+  // IMPORTANT: the returned bounding boxes are expressed in the unrotated input
+  // frame of reference coordinates system, i.e. in `[0, frame_buffer.width) x
+  // [0, frame_buffer.height)`, which are the dimensions of the underlying
+  // `frame_buffer` data before any `Orientation` flag gets applied.
+  //
+  // In particular, this implies that the returned bounding boxes may not be
+  // directly suitable for display if the input image is displayed *with* the
+  // `Orientation` flag taken into account according to the EXIF specification
+  // (http://jpegclub.org/exif_orientation.html): it may first need to be
+  // rotated. This is typically true when consuming camera frames on Android or
+  // iOS.
+  //
+  // For example, if the input `frame_buffer` has its `Orientation` flag set to
+  // `kLeftBottom` (i.e. the image will be rotated 90° clockwise during
+  // preprocessing to make it "upright"), then the same 90° clockwise rotation
+  // needs to be applied to the bounding box for display.
+  tflite::support::StatusOr<DetectionResult> Detect(
+      const FrameBuffer& frame_buffer);
+
+ protected:
+  // Post-processing to transform the raw model outputs into detection results.
+  tflite::support::StatusOr<DetectionResult> Postprocess(
+      const std::vector<const TfLiteTensor*>& output_tensors,
+      const FrameBuffer& frame_buffer, const BoundingBox& roi) override;
+
+  // Performs sanity checks on the provided ObjectDetectorOptions.
+  static absl::Status SanityCheckOptions(const ObjectDetectorOptions& options);
+
+  // Initializes the ObjectDetector from the provided ObjectDetectorOptions,
+  // whose ownership is transferred to this object.
+  absl::Status Init(std::unique_ptr<ObjectDetectorOptions>);
+
+  // Performs pre-initialization actions.
+  virtual absl::Status PreInit();
+
+ private:
+  // Performs sanity checks on the model outputs and extracts their metadata.
+  absl::Status CheckAndSetOutputs();
+
+  // Performs sanity checks on the class whitelist/blacklist and forms the class
+  // index set.
+  absl::Status CheckAndSetClassIndexSet();
+
+  // Checks if the class at the provided index is allowed, i.e. whitelisted in
+  // case a whitelist is provided or not blacklisted if a blacklist is provided.
+  // Always returns true if no whitelist or blacklist were provided.
+  bool IsClassIndexAllowed(int class_index);
+
+  // Given a DetectionResult object containing class indices, fills the name and
+  // display name from the label map.
+  absl::Status FillResultsFromLabelMap(DetectionResult* result);
+
+  // The options used to build this ObjectDetector.
+  std::unique_ptr<ObjectDetectorOptions> options_;
+
+  // This is populated by reading the label files from the TFLite Model
+  // Metadata: if no such files are available, this is left empty and the
+  // ObjectDetector will only be able to populate the `index` field of the
+  // detection results `classes` field.
+  std::vector<LabelMapItem> label_map_;
+
+  // For each pack of 4 coordinates returned by the model, this denotes the
+  // order in which to get the left, top, right and bottom coordinates.
+  std::vector<unsigned int> bounding_box_corners_order_;
+
+  // Set of whitelisted or blacklisted class indices.
+  struct ClassIndexSet {
+    absl::flat_hash_set<int> values;
+    bool is_whitelist;
+  };
+  // Whitelisted or blacklisted class indices based on provided options at
+  // construction time. These are used to filter out results during
+  // post-processing.
+  ClassIndexSet class_index_set_;
+
+  // Score threshold. Detections with a confidence below this value are
+  // discarded. If none is provided via metadata or options, -FLT_MAX is set as
+  // default value.
+  float score_threshold_;
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_OBJECT_DETECTOR_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/BUILD b/tensorflow_lite_support/cc/task/vision/proto/BUILD
new file mode 100644
index 0000000..e294da7
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/BUILD
@@ -0,0 +1,208 @@
+load("//tensorflow_lite_support/cc/port:build_defs.bzl", "support_cc_proto_library")
+
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+# Common vision protos.
+
+proto_library(
+    name = "bounding_box_proto",
+    srcs = ["bounding_box.proto"],
+)
+
+support_cc_proto_library(
+    name = "bounding_box_cc_proto",
+    srcs = ["bounding_box.proto"],
+    deps = [
+        ":bounding_box_proto",
+    ],
+)
+
+cc_library(
+    name = "bounding_box_proto_inc",
+    hdrs = ["bounding_box_proto_inc.h"],
+    deps = [":bounding_box_cc_proto"],
+)
+
+proto_library(
+    name = "class_proto",
+    srcs = ["class.proto"],
+)
+
+support_cc_proto_library(
+    name = "class_cc_proto",
+    srcs = ["class.proto"],
+    deps = [
+        ":class_proto",
+    ],
+)
+
+cc_library(
+    name = "class_proto_inc",
+    hdrs = ["class_proto_inc.h"],
+    deps = [":class_cc_proto"],
+)
+
+# ObjectDetector protos.
+
+proto_library(
+    name = "object_detector_options_proto",
+    srcs = ["object_detector_options.proto"],
+    deps = [
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto",
+    ],
+)
+
+support_cc_proto_library(
+    name = "object_detector_options_cc_proto",
+    srcs = ["object_detector_options.proto"],
+    cc_deps = ["//tensorflow_lite_support/cc/task/core/proto:external_file_cc_proto"],
+    deps = [
+        ":object_detector_options_proto",
+    ],
+)
+
+cc_library(
+    name = "object_detector_options_proto_inc",
+    hdrs = ["object_detector_options_proto_inc.h"],
+    deps = [
+        ":object_detector_options_cc_proto",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+    ],
+)
+
+proto_library(
+    name = "detections_proto",
+    srcs = ["detections.proto"],
+    deps = [
+        ":bounding_box_proto",
+        ":class_proto",
+    ],
+)
+
+support_cc_proto_library(
+    name = "detections_cc_proto",
+    srcs = ["detections.proto"],
+    cc_deps = [
+        ":bounding_box_cc_proto",
+        ":class_cc_proto",
+    ],
+    deps = [
+        ":detections_proto",
+    ],
+)
+
+cc_library(
+    name = "detections_proto_inc",
+    hdrs = ["detections_proto_inc.h"],
+    deps = [
+        ":bounding_box_proto_inc",
+        ":class_proto_inc",
+        ":detections_cc_proto",
+    ],
+)
+
+# ImageClassifier protos.
+
+proto_library(
+    name = "image_classifier_options_proto",
+    srcs = ["image_classifier_options.proto"],
+    deps = [
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto",
+    ],
+)
+
+support_cc_proto_library(
+    name = "image_classifier_options_cc_proto",
+    srcs = ["image_classifier_options.proto"],
+    cc_deps = ["//tensorflow_lite_support/cc/task/core/proto:external_file_cc_proto"],
+    deps = [
+        ":image_classifier_options_proto",
+    ],
+)
+
+cc_library(
+    name = "image_classifier_options_proto_inc",
+    hdrs = ["image_classifier_options_proto_inc.h"],
+    deps = [
+        ":image_classifier_options_cc_proto",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+    ],
+)
+
+proto_library(
+    name = "classifications_proto",
+    srcs = ["classifications.proto"],
+    deps = [
+        ":class_proto",
+    ],
+)
+
+support_cc_proto_library(
+    name = "classifications_cc_proto",
+    srcs = ["classifications.proto"],
+    cc_deps = [":class_cc_proto"],
+    deps = [
+        ":classifications_proto",
+    ],
+)
+
+cc_library(
+    name = "classifications_proto_inc",
+    hdrs = ["classifications_proto_inc.h"],
+    deps = [
+        ":class_proto_inc",
+        ":classifications_cc_proto",
+    ],
+)
+
+# ImageSegmenter protos.
+
+proto_library(
+    name = "image_segmenter_options_proto",
+    srcs = ["image_segmenter_options.proto"],
+    deps = [
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto",
+    ],
+)
+
+support_cc_proto_library(
+    name = "image_segmenter_options_cc_proto",
+    srcs = ["image_segmenter_options.proto"],
+    cc_deps = ["//tensorflow_lite_support/cc/task/core/proto:external_file_cc_proto"],
+    deps = [
+        ":image_segmenter_options_proto",
+    ],
+)
+
+cc_library(
+    name = "image_segmenter_options_proto_inc",
+    hdrs = ["image_segmenter_options_proto_inc.h"],
+    deps = [
+        ":image_segmenter_options_cc_proto",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+    ],
+)
+
+proto_library(
+    name = "segmentations_proto",
+    srcs = ["segmentations.proto"],
+)
+
+support_cc_proto_library(
+    name = "segmentations_cc_proto",
+    srcs = ["segmentations.proto"],
+    deps = [
+        ":segmentations_proto",
+    ],
+)
+
+cc_library(
+    name = "segmentations_proto_inc",
+    hdrs = ["segmentations_proto_inc.h"],
+    deps = [":segmentations_cc_proto"],
+)
diff --git a/tensorflow_lite_support/cc/task/vision/proto/bounding_box.proto b/tensorflow_lite_support/cc/task/vision/proto/bounding_box.proto
new file mode 100644
index 0000000..4c2e130
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/bounding_box.proto
@@ -0,0 +1,30 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+// An integer bounding box, axis aligned.
+message BoundingBox {
+  // The X coordinate of the top-left corner, in pixels.
+  optional int32 origin_x = 1;
+  // The Y coordinate of the top-left corner, in pixels.
+  optional int32 origin_y = 2;
+  // The width of the bounding box, in pixels.
+  optional int32 width = 3;
+  // The height of the bounding box, in pixels.
+  optional int32 height = 4;
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h
new file mode 100644
index 0000000..ef84b15
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h
@@ -0,0 +1,19 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_BOUNDING_BOX_PROTO_INC_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_BOUNDING_BOX_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box.pb.h"
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_BOUNDING_BOX_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/class.proto b/tensorflow_lite_support/cc/task/vision/proto/class.proto
new file mode 100644
index 0000000..19e8ac1
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/class.proto
@@ -0,0 +1,36 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+// A single classification result.
+message Class {
+  // The index of the class in the corresponding label map, usually packed in
+  // the TFLite Model Metadata [1].
+  //
+  // [1]: https://www.tensorflow.org/lite/convert/metadata
+  optional int32 index = 1;
+  // The score for this class e.g. (but not necessarily) a probability in [0,1].
+  optional float score = 2;
+  // A human readable name of the class filled from the label map.
+  optional string display_name = 3;
+  // An ID for the class, not necessarily human-readable (e.g. a Google
+  // Knowledge Graph ID [1]), filled from the label map.
+  //
+  // [1]: https://developers.google.com/knowledge-graph
+  optional string class_name = 4;
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h
new file mode 100644
index 0000000..2f9a409
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h
@@ -0,0 +1,20 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_CLASS_PROTO_INC_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_CLASS_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/vision/proto/class.pb.h"
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_CLASS_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/classifications.proto b/tensorflow_lite_support/cc/task/vision/proto/classifications.proto
new file mode 100644
index 0000000..d3d9c66
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/classifications.proto
@@ -0,0 +1,35 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+import "tensorflow_lite_support/cc/task/vision/proto/class.proto";
+
+// List of predicted classes (aka labels) for a given image classifier head.
+message Classifications {
+  // The array of predicted classes, usually sorted by descending scores (e.g.
+  // from high to low probability).
+  repeated Class classes = 1;
+  // The index of the image classifier head these classes refer to. This is
+  // useful for multi-head models.
+  optional int32 head_index = 2;
+}
+
+// Contains one set of results per image classifier head.
+message ClassificationResult {
+  repeated Classifications classifications = 1;
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/classifications_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/classifications_proto_inc.h
new file mode 100644
index 0000000..62a5f11
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/classifications_proto_inc.h
@@ -0,0 +1,22 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_CLASSIFICATIONS_PROTO_INC_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_CLASSIFICATIONS_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h"
+
+#include "tensorflow_lite_support/cc/task/vision/proto/classifications.pb.h"
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_CLASSIFICATIONS_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/detections.proto b/tensorflow_lite_support/cc/task/vision/proto/detections.proto
new file mode 100644
index 0000000..b600fc9
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/detections.proto
@@ -0,0 +1,53 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+import "tensorflow_lite_support/cc/task/vision/proto/bounding_box.proto";
+import "tensorflow_lite_support/cc/task/vision/proto/class.proto";
+
+// A single detected object.
+message Detection {
+  // The bounding box.
+  //
+  // IMPORTANT: when using the Task APIs, the bounding box is expressed in the
+  // unrotated input frame of reference coordinates system, i.e. in `[0,
+  // frame_buffer.width) x [0, frame_buffer.height)`, which are the dimensions
+  // of the underlying `frame_buffer` data before any `Orientation` flag gets
+  // applied.
+  //
+  // In particular, this implies that the returned bounding boxes may not be
+  // directly suitable for display if the input image is displayed *with* the
+  // `Orientation` flag taken into account according to the EXIF specification
+  // (http://jpegclub.org/exif_orientation.html): it may first need to be
+  // rotated.
+  //
+  // For example, if the input `frame_buffer` has its `Orientation` flag set to
+  // `kLeftBottom` (i.e. the image will be rotated 90° clockwise during
+  // preprocessing to make it "upright"), then the same 90° clockwise rotation
+  // needs to be applied to the bounding box for display.
+  optional BoundingBox bounding_box = 2;
+  // The candidate classes, sorted by descending score.
+  repeated Class classes = 3;
+  // Reserved tags.
+  reserved 1, 4;
+}
+
+// List of detected objects.
+message DetectionResult {
+  repeated Detection detections = 1;
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/detections_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/detections_proto_inc.h
new file mode 100644
index 0000000..2b63cad
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/detections_proto_inc.h
@@ -0,0 +1,23 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_DETECTIONS_PROTO_INC_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_DETECTIONS_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h"
+
+#include "tensorflow_lite_support/cc/task/vision/proto/detections.pb.h"
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_DETECTIONS_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/image_classifier_options.proto b/tensorflow_lite_support/cc/task/vision/proto/image_classifier_options.proto
new file mode 100644
index 0000000..24cd85f
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/image_classifier_options.proto
@@ -0,0 +1,67 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+import "tensorflow_lite_support/cc/task/core/proto/external_file.proto";
+
+// Options for setting up an ImageClassifier.
+// Next Id: 14
+message ImageClassifierOptions {
+  // The external model file, as a single standalone TFLite file. If it is
+  // packed with TFLite Model Metadata [1], those are used to populate e.g. the
+  // label map, score calibration and recommended score thresholds. Models
+  // without any such metadata or partial metadata are supported, but may result
+  // in the image classifier providing degraded functionality; typically, a
+  // model that doesn't contain any label map won't be able to return any class
+  // or display names but will be limited to returning class indices.
+  //
+  // [1]: https://www.tensorflow.org/lite/convert/metadata
+  optional core.ExternalFile model_file_with_metadata = 10;
+
+  // The locale to use for display names specified through the TFLite Model
+  // Metadata, if any. Defaults to English.
+  optional string display_names_locale = 11 [default = "en"];
+
+  // The maximum number of top-scored classification results to return. If < 0,
+  // all available results will be returned. If 0, an invalid argument error is
+  // returned.
+  optional int32 max_results = 2 [default = -1];
+
+  // Score threshold in [0,1), overrides the ones provided in the model metadata
+  // (if any). Results below this value are rejected.
+  optional float score_threshold = 3;
+
+  // Optional whitelist of class names. If non-empty, classifications whose
+  // class name is not in this set will be filtered out. Duplicate or unknown
+  // class names are ignored. Mutually exclusive with class_name_blacklist.
+  repeated string class_name_whitelist = 4;
+
+  // Optional blacklist of class names. If non-empty, classifications whose
+  // class name is in this set will be filtered out. Duplicate or unknown
+  // class names are ignored. Mutually exclusive with class_name_whitelist.
+  repeated string class_name_blacklist = 5;
+
+  // The number of threads to be used for TFLite ops that support
+  // multi-threading when running inference with CPU.
+  // num_threads should be greater than 0 or equal to -1. Setting num_threads to
+  // -1 has the effect to let TFLite runtime set the value.
+  optional int32 num_threads = 13 [default = -1];
+
+  // Reserved tags.
+  reserved 1, 6, 7, 8, 9, 12;
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/image_classifier_options_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/image_classifier_options_proto_inc.h
new file mode 100644
index 0000000..03dcd75
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/image_classifier_options_proto_inc.h
@@ -0,0 +1,22 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_CLASSIFIER_OPTIONS_PROTO_INC_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_CLASSIFIER_OPTIONS_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+
+#include "tensorflow_lite_support/cc/task/vision/proto/image_classifier_options.pb.h"
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_CLASSIFIER_OPTIONS_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options.proto b/tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options.proto
new file mode 100644
index 0000000..3afed86
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options.proto
@@ -0,0 +1,61 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+import "tensorflow_lite_support/cc/task/core/proto/external_file.proto";
+
+// Options for setting up an ImageSegmenter.
+// Next Id: 8
+message ImageSegmenterOptions {
+  // The external model file, as a single standalone TFLite file. If it is
+  // packed with TFLite Model Metadata [1], those are used to populate label
+  // map. Models without any such metadata or partial metadata are supported,
+  // but may result in the segmenter providing degraded functionality;
+  // typically, a model that doesn't contain any label map won't be able to
+  // return any class or display names.
+  //
+  // [1]: https://www.tensorflow.org/lite/convert/metadata
+  optional core.ExternalFile model_file_with_metadata = 5;
+
+  // The locale to use for display names specified through the TFLite Model
+  // Metadata, if any. Defaults to English.
+  optional string display_names_locale = 6 [default = "en"];
+
+  // Output mask type. This allows specifying the type of post-processing to
+  // perform on the raw model results (see SegmentationResult proto for more).
+  enum OutputType {
+    UNSPECIFIED = 0;
+    // Gives a single output mask where each pixel represents the class which
+    // the pixel in the original image was predicted to belong to.
+    CATEGORY_MASK = 1;
+    // Gives a list of output masks where, for each mask, each pixel represents
+    // the prediction confidence, usually in the [0, 1] range.
+    CONFIDENCE_MASK = 2;
+  }
+  // Optional output mask type.
+  optional OutputType output_type = 3 [default = CATEGORY_MASK];
+
+  // The number of threads to be used for TFLite ops that support
+  // multi-threading when running inference with CPU.
+  // num_threads should be greater than 0 or equal to -1. Setting num_threads to
+  // -1 has the effect to let TFLite runtime set the value.
+  optional int32 num_threads = 7 [default = -1];
+
+  // Reserved tags.
+  reserved 1, 2, 4;
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options_proto_inc.h
new file mode 100644
index 0000000..aaaecf3
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options_proto_inc.h
@@ -0,0 +1,22 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_SEGMENTER_OPTIONS_PROTO_INC_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_SEGMENTER_OPTIONS_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+
+#include "tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options.pb.h"
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_SEGMENTER_OPTIONS_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/object_detector_options.proto b/tensorflow_lite_support/cc/task/vision/proto/object_detector_options.proto
new file mode 100644
index 0000000..b55e974
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/object_detector_options.proto
@@ -0,0 +1,62 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+import "tensorflow_lite_support/cc/task/core/proto/external_file.proto";
+
+// Options for setting up an ObjectDetector.
+// Next Id: 8.
+message ObjectDetectorOptions {
+  // The external model file, as a single standalone TFLite file packed with
+  // TFLite Model Metadata [1]. Those are mandatory, and used to populate e.g.
+  // the label map and recommended score threshold.
+  //
+  // [1]: https://www.tensorflow.org/lite/convert/metadata
+  optional core.ExternalFile model_file_with_metadata = 1;
+
+  // The locale to use for display names specified through the TFLite Model
+  // Metadata, if any. Defaults to English.
+  optional string display_names_locale = 2 [default = "en"];
+
+  // The maximum number of top-scored detection results to return. If < 0, all
+  // available results will be returned. If 0, an invalid argument error is
+  // returned. Note that models may intrinsically be limited to returning a
+  // maximum number of results N: if the provided value here is above N, only N
+  // results will be returned.
+  optional int32 max_results = 3 [default = -1];
+
+  // Score threshold to override the one provided in the model metadata (if
+  // any). Detection results with a score below this value are rejected.
+  optional float score_threshold = 4;
+
+  // Optional whitelist of class names. If non-empty, detection results whose
+  // class name is not in this set will be filtered out. Duplicate or unknown
+  // class names are ignored. Mutually exclusive with class_name_blacklist.
+  repeated string class_name_whitelist = 5;
+
+  // Optional blacklist of class names. If non-empty, detection results whose
+  // class name is in this set will be filtered out. Duplicate or unknown
+  // class names are ignored. Mutually exclusive with class_name_whitelist.
+  repeated string class_name_blacklist = 6;
+
+  // The number of threads to be used for TFLite ops that support
+  // multi-threading when running inference with CPU.
+  // num_threads should be greater than 0 or equal to -1. Setting num_threads to
+  // -1 has the effect to let TFLite runtime set the value.
+  optional int32 num_threads = 7 [default = -1];
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/object_detector_options_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/object_detector_options_proto_inc.h
new file mode 100644
index 0000000..2789847
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/object_detector_options_proto_inc.h
@@ -0,0 +1,22 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_OBJECT_DETECTOR_OPTIONS_PROTO_INC_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_OBJECT_DETECTOR_OPTIONS_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+
+#include "tensorflow_lite_support/cc/task/vision/proto/object_detector_options.pb.h"
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_OBJECT_DETECTOR_OPTIONS_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/proto/segmentations.proto b/tensorflow_lite_support/cc/task/vision/proto/segmentations.proto
new file mode 100644
index 0000000..259bee8
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/segmentations.proto
@@ -0,0 +1,109 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+syntax = "proto2";
+
+package tflite.task.vision;
+
+// Results of performing image segmentation.
+// Note that at the time, a single `Segmentation` element is expected to be
+// returned; the field is made repeated for later extension to e.g. instance
+// segmentation models, which may return one segmentation per object.
+message SegmentationResult {
+  repeated Segmentation segmentation = 1;
+}
+
+// Next Id: 6
+message Segmentation {
+  // Confidence mask. This is a flattened 2D-array in row major order. For each
+  // pixel, the value indicates the prediction confidence usually in the [0, 1]
+  // range where higher values represent a stronger confidence. Ultimately this
+  // is model specific, and other range of values might be used.
+  message ConfidenceMask {
+    repeated float value = 1 [packed = true];
+  }
+
+  // List of confidence masks with respect to the model output depth (this depth
+  // represents how many classes are supported). Note: some models have a single
+  // class (e.g. a sky segmentation model) which turns into a single confidence
+  // mask in this list.
+  message ConfidenceMasks {
+    repeated ConfidenceMask confidence_mask = 1;
+  }
+
+  // IMPORTANT: segmentation masks are not direcly suited for display, in
+  // particular:
+  // * they are relative to the unrotated input frame, i.e. *not* taking into
+  //   account the `Orientation` flag of the input FrameBuffer,
+  // * their dimensions are intrinsic to the model, i.e. *not* dependent on the
+  //   input FrameBuffer dimensions.
+  //
+  // Example of such post-processing, assuming:
+  // * an input FrameBuffer with width=640, height=480, orientation=kLeftBottom
+  //   (i.e. the image will be rotated 90° clockwise during preprocessing to
+  //   make it "upright"),
+  // * a model outputting masks of size 224x224.
+  // In order to be directly displayable on top of the input image assumed to
+  // be displayed *with* the `Orientation` flag taken into account (according to
+  // the EXIF specification [1]), the masks need to be:
+  // * re-scaled to 640 x 480,
+  // * then rotated 90° clockwise.
+  //
+  // [1]: http://jpegclub.org/exif_orientation.html
+  oneof mask_oneof {
+    // Category mask. This is a flattened 2D-array of size `width` x `height`,
+    // in row major order. The value of each pixel in this mask represents the
+    // class to which the pixel belongs.
+    // See `colored_labels` for instructions on how to get pixel labels and
+    // display color.
+    bytes category_mask = 1;
+
+    // One confidence masks of size `width` x `height` for each of the supported
+    // classes. The value of each pixel in these masks represents the confidence
+    // score for this particular class.
+    // See `colored_labels` for instructions on how to get pixel labels and
+    // display color.
+    ConfidenceMasks confidence_masks = 4;
+  }
+  // The width of the mask. This is an intrinsic parameter of the model being
+  // used, and does not depend on the input image dimensions.
+  optional int32 width = 2;
+  // The height of the mask. This is an intrinsic parameter of the model being
+  // used, and does not depend on the input image dimensions.
+  optional int32 height = 3;
+
+  // Defines a label associated with an RGB color, for display purposes.
+  message ColoredLabel {
+    // The RGB color components for the label, in the [0, 255] range.
+    optional uint32 r = 1;
+    optional uint32 g = 2;
+    optional uint32 b = 3;
+    // The class name, as provided in the label map packed in the TFLite Model
+    // Metadata.
+    optional string class_name = 4;
+    // The display name, as provided in the label map (if available) packed in
+    // the TFLite Model Metadata. See `display_names_locale` field in
+    // ImageSegmenterOptions.
+    optional string display_name = 5;
+  }
+
+  // The list of colored labels for all the supported categories. Depending on
+  // which is present, this list is in 1:1 correspondence with:
+  // * `category_mask` pixel values, i.e. a pixel with value `i` is
+  //   associated with `colored_labels[i]`,
+  // * `confidence_masks` indices, i.e. `confidence_masks[i]` is associated with
+  //   `colored_labels[i]`.
+  repeated ColoredLabel colored_labels = 5;
+}
diff --git a/tensorflow_lite_support/cc/task/vision/proto/segmentations_proto_inc.h b/tensorflow_lite_support/cc/task/vision/proto/segmentations_proto_inc.h
new file mode 100644
index 0000000..cfc96e6
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/proto/segmentations_proto_inc.h
@@ -0,0 +1,19 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_SEGMENTATIONS_PROTO_INC_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_SEGMENTATIONS_PROTO_INC_H_
+
+#include "tensorflow_lite_support/cc/task/vision/proto/segmentations.pb.h"
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_SEGMENTATIONS_PROTO_INC_H_
diff --git a/tensorflow_lite_support/cc/task/vision/utils/BUILD b/tensorflow_lite_support/cc/task/vision/utils/BUILD
new file mode 100644
index 0000000..8995145
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/BUILD
@@ -0,0 +1,109 @@
+package(
+    default_visibility = [
+        "//tensorflow_lite_support:users",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "score_calibration",
+    srcs = ["score_calibration.cc"],
+    hdrs = ["score_calibration.h"],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision/core:label_map_item",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "@com_google_absl//absl/container:flat_hash_map",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@com_google_absl//absl/types:optional",
+    ],
+)
+
+cc_library(
+    name = "frame_buffer_common_utils",
+    srcs = [
+        "frame_buffer_common_utils.cc",
+    ],
+    hdrs = [
+        "frame_buffer_common_utils.h",
+        "frame_buffer_utils_interface.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@com_google_absl//absl/time",
+    ],
+)
+
+cc_library(
+    name = "frame_buffer_utils",
+    srcs = [
+        "frame_buffer_utils.cc",
+    ],
+    hdrs = [
+        "frame_buffer_utils.h",
+    ],
+    deps = [
+        ":frame_buffer_common_utils",
+        ":libyuv_frame_buffer_utils",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@com_google_absl//absl/types:optional",
+        "@com_google_absl//absl/types:variant",
+        "@org_tensorflow//tensorflow/lite/kernels:op_macros",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:compatibility",
+    ],
+)
+
+cc_library(
+    name = "libyuv_frame_buffer_utils",
+    srcs = ["libyuv_frame_buffer_utils.cc"],
+    hdrs = ["libyuv_frame_buffer_utils.h"],
+    deps = [
+        ":frame_buffer_common_utils",
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@libyuv",
+    ],
+)
+
+cc_library(
+    name = "image_tensor_specs",
+    srcs = ["image_tensor_specs.cc"],
+    hdrs = ["image_tensor_specs.h"],
+    deps = [
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:tflite_engine",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/types:optional",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/c:common",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.cc b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.cc
new file mode 100644
index 0000000..fa9b05f
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.cc
@@ -0,0 +1,428 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+
+#include <string>
+#include <vector>
+
+#include "absl/strings/str_cat.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+namespace {
+
+using ::tflite::support::StatusOr;
+
+constexpr int kRgbaChannels = 4;
+constexpr int kRgbChannels = 3;
+constexpr int kGrayChannel = 1;
+
+// Creates a FrameBuffer from raw NV12 buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromNV12RawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation, const absl::Time timestamp) {
+  const std::vector<FrameBuffer::Plane> planes_nv12 = {
+      {input, /*stride=*/{dimension.width, kGrayChannel}},
+      {input + dimension.Size(), /*stride=*/{dimension.width, 2}}};
+  return FrameBuffer::Create(planes_nv12, dimension, FrameBuffer::Format::kNV12,
+                             orientation, timestamp);
+}
+
+// Creates a FrameBuffer from raw NV21 buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromNV21RawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation, const absl::Time timestamp) {
+  FrameBuffer::Plane input_plane = {/*buffer=*/input,
+                                    /*stride=*/{dimension.width, kGrayChannel}};
+  return FrameBuffer::Create({input_plane}, dimension,
+                             FrameBuffer::Format::kNV21, orientation,
+                             timestamp);
+}
+
+// Indicates whether the given buffers have the same dimensions.
+bool AreBufferDimsEqual(const FrameBuffer& buffer1,
+                        const FrameBuffer& buffer2) {
+  return buffer1.dimension() == buffer2.dimension();
+}
+
+// Indicates whether the given buffers formats are compatible. Same formats are
+// compatible and all YUV family formats (e.g. NV21, NV12, YV12, YV21, etc) are
+// compatible.
+bool AreBufferFormatsCompatible(const FrameBuffer& buffer1,
+                                const FrameBuffer& buffer2) {
+  switch (buffer1.format()) {
+    case FrameBuffer::Format::kRGBA:
+    case FrameBuffer::Format::kRGB:
+      return (buffer2.format() == FrameBuffer::Format::kRGBA ||
+              buffer2.format() == FrameBuffer::Format::kRGB);
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return (buffer2.format() == FrameBuffer::Format::kNV12 ||
+              buffer2.format() == FrameBuffer::Format::kNV21 ||
+              buffer2.format() == FrameBuffer::Format::kYV12 ||
+              buffer2.format() == FrameBuffer::Format::kYV21);
+    case FrameBuffer::Format::kGRAY:
+    default:
+      return buffer1.format() == buffer2.format();
+  }
+}
+
+}  // namespace
+
+// Miscellaneous Methods
+// -----------------------------------------------------------------
+int GetFrameBufferByteSize(FrameBuffer::Dimension dimension,
+                           FrameBuffer::Format format) {
+  switch (format) {
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return /*y plane*/ dimension.Size() +
+             /*uv plane*/ ((static_cast<float>(dimension.width + 1) / 2) *
+                           (static_cast<float>(dimension.height + 1) / 2) * 2);
+    case FrameBuffer::Format::kRGB:
+      return dimension.Size() * 3;
+    case FrameBuffer::Format::kRGBA:
+      return dimension.Size() * 4;
+    case FrameBuffer::Format::kGRAY:
+      return dimension.Size();
+    default:
+      return 0;
+  }
+}
+
+StatusOr<int> GetPixelStrides(FrameBuffer::Format format) {
+  switch (format) {
+    case FrameBuffer::Format::kGRAY:
+      return kGrayPixelBytes;
+    case FrameBuffer::Format::kRGB:
+      return kRgbPixelBytes;
+    case FrameBuffer::Format::kRGBA:
+      return kRgbaPixelBytes;
+    default:
+      return absl::InvalidArgumentError(absl::StrFormat(
+          "GetPixelStrides does not support format: %i.", format));
+  }
+}
+
+StatusOr<const uint8*> GetUvRawBuffer(const FrameBuffer& buffer) {
+  if (buffer.format() != FrameBuffer::Format::kNV12 &&
+      buffer.format() != FrameBuffer::Format::kNV21) {
+    return absl::InvalidArgumentError(
+        "Only support getting biplanar UV buffer from NV12/NV21 frame buffer.");
+  }
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData yuv_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  const uint8* uv_buffer = buffer.format() == FrameBuffer::Format::kNV12
+                               ? yuv_data.u_buffer
+                               : yuv_data.v_buffer;
+  return uv_buffer;
+}
+
+StatusOr<FrameBuffer::Dimension> GetUvPlaneDimension(
+    FrameBuffer::Dimension dimension, FrameBuffer::Format format) {
+  if (dimension.width <= 0 || dimension.height <= 0) {
+    return absl::InvalidArgumentError(
+        absl::StrFormat("Invalid input dimension: {%d, %d}.", dimension.width,
+                        dimension.height));
+  }
+  switch (format) {
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return FrameBuffer::Dimension{(dimension.width + 1) / 2,
+                                    (dimension.height + 1) / 2};
+    default:
+      return absl::InvalidArgumentError(
+          absl::StrFormat("Input format is not YUV-like: %i.", format));
+  }
+}
+
+FrameBuffer::Dimension GetCropDimension(int x0, int x1, int y0, int y1) {
+  return {x1 - x0 + 1, y1 - y0 + 1};
+}
+
+// Validation Methods
+// -----------------------------------------------------------------
+
+absl::Status ValidateBufferPlaneMetadata(const FrameBuffer& buffer) {
+  if (buffer.plane_count() < 1) {
+    return absl::InvalidArgumentError(
+        "There must be at least 1 plane specified.");
+  }
+
+  for (int i = 0; i < buffer.plane_count(); i++) {
+    if (buffer.plane(i).stride.row_stride_bytes == 0 ||
+        buffer.plane(i).stride.pixel_stride_bytes == 0) {
+      return absl::InvalidArgumentError("Invalid stride information.");
+    }
+  }
+
+  return absl::OkStatus();
+}
+
+absl::Status ValidateBufferFormat(const FrameBuffer& buffer) {
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kGRAY:
+    case FrameBuffer::Format::kRGB:
+    case FrameBuffer::Format::kRGBA:
+      if (buffer.plane_count() == 1) return absl::OkStatus();
+      return absl::InvalidArgumentError(
+          "Plane count must be 1 for grayscale and RGB[a] buffers.");
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kYV21:
+    case FrameBuffer::Format::kYV12:
+      return absl::OkStatus();
+    default:
+      return absl::InternalError(
+          absl::StrFormat("Unsupported buffer format: %i.", buffer.format()));
+  }
+}
+
+absl::Status ValidateBufferFormats(const FrameBuffer& buffer1,
+                                   const FrameBuffer& buffer2) {
+  RETURN_IF_ERROR(ValidateBufferFormat(buffer1));
+  RETURN_IF_ERROR(ValidateBufferFormat(buffer2));
+  return absl::OkStatus();
+}
+
+absl::Status ValidateResizeBufferInputs(const FrameBuffer& buffer,
+                                        const FrameBuffer& output_buffer) {
+  bool valid_format = false;
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kGRAY:
+    case FrameBuffer::Format::kRGB:
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      valid_format = (buffer.format() == output_buffer.format());
+      break;
+    case FrameBuffer::Format::kRGBA:
+      valid_format = (output_buffer.format() == FrameBuffer::Format::kRGBA ||
+                      output_buffer.format() == FrameBuffer::Format::kRGB);
+      break;
+    default:
+      return absl::InternalError(
+          absl::StrFormat("Unsupported buffer format: %i.", buffer.format()));
+  }
+  if (!valid_format) {
+    return absl::InvalidArgumentError(
+        "Input and output buffer formats must match.");
+  }
+  return ValidateBufferFormats(buffer, output_buffer);
+}
+
+absl::Status ValidateRotateBufferInputs(const FrameBuffer& buffer,
+                                        const FrameBuffer& output_buffer,
+                                        int angle_deg) {
+  if (!AreBufferFormatsCompatible(buffer, output_buffer)) {
+    return absl::InvalidArgumentError(
+        "Input and output buffer formats must match.");
+  }
+
+  const bool is_dimension_change = (angle_deg / 90) % 2 == 1;
+  const bool are_dimensions_rotated =
+      (buffer.dimension().width == output_buffer.dimension().height) &&
+      (buffer.dimension().height == output_buffer.dimension().width);
+  const bool are_dimensions_equal =
+      buffer.dimension() == output_buffer.dimension();
+
+  if (angle_deg >= 360 || angle_deg <= 0 || angle_deg % 90 != 0) {
+    return absl::InvalidArgumentError(
+        "Rotation angle must be between 0 and 360, in multiples of 90 "
+        "degrees.");
+  } else if ((is_dimension_change && !are_dimensions_rotated) ||
+             (!is_dimension_change && !are_dimensions_equal)) {
+    return absl::InvalidArgumentError(
+        "Output buffer has invalid dimensions for rotation.");
+  }
+  return absl::OkStatus();
+}
+
+absl::Status ValidateCropBufferInputs(const FrameBuffer& buffer,
+                                      const FrameBuffer& output_buffer, int x0,
+                                      int y0, int x1, int y1) {
+  if (!AreBufferFormatsCompatible(buffer, output_buffer)) {
+    return absl::InvalidArgumentError(
+        "Input and output buffer formats must match.");
+  }
+
+  bool is_buffer_size_valid =
+      ((x1 < buffer.dimension().width) && y1 < buffer.dimension().height);
+  bool are_points_valid = (x0 >= 0) && (y0 >= 0) && (x1 >= x0) && (y1 >= y0);
+
+  if (!is_buffer_size_valid || !are_points_valid) {
+    return absl::InvalidArgumentError("Invalid crop coordinates.");
+  }
+  return absl::OkStatus();
+}
+
+absl::Status ValidateFlipBufferInputs(const FrameBuffer& buffer,
+                                      const FrameBuffer& output_buffer) {
+  if (!AreBufferFormatsCompatible(buffer, output_buffer)) {
+    return absl::InvalidArgumentError(
+        "Input and output buffer formats must match.");
+  }
+  return AreBufferDimsEqual(buffer, output_buffer)
+             ? absl::OkStatus()
+             : absl::InvalidArgumentError(
+                   "Input and output buffers must have the same dimensions.");
+}
+
+absl::Status ValidateConvertFormats(FrameBuffer::Format from_format,
+                                    FrameBuffer::Format to_format) {
+  if (from_format == to_format) {
+    return absl::InvalidArgumentError("Formats must be different.");
+  }
+
+  switch (from_format) {
+    case FrameBuffer::Format::kGRAY:
+      return absl::InvalidArgumentError(
+          "Grayscale format does not convert to other formats.");
+    case FrameBuffer::Format::kRGB:
+      if (to_format == FrameBuffer::Format::kRGBA) {
+        return absl::InvalidArgumentError(
+            "RGB format does not convert to RGBA");
+      }
+      return absl::OkStatus();
+    case FrameBuffer::Format::kRGBA:
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return absl::OkStatus();
+    default:
+      return absl::InternalError(
+          absl::StrFormat("Unsupported buffer format: %i.", from_format));
+  }
+}
+
+// Creation Methods
+// -----------------------------------------------------------------
+
+// Creates a FrameBuffer from raw RGBA buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromRgbaRawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation, const absl::Time timestamp) {
+  FrameBuffer::Plane input_plane = {
+      /*buffer=*/input,
+      /*stride=*/{dimension.width * kRgbaChannels, kRgbaChannels}};
+  return FrameBuffer::Create({input_plane}, dimension,
+                             FrameBuffer::Format::kRGBA, orientation,
+                             timestamp);
+}
+
+// Creates a FrameBuffer from raw RGB buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromRgbRawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation, const absl::Time timestamp) {
+  FrameBuffer::Plane input_plane = {
+      /*buffer=*/input,
+      /*stride=*/{dimension.width * kRgbChannels, kRgbChannels}};
+  return FrameBuffer::Create({input_plane}, dimension,
+                             FrameBuffer::Format::kRGB, orientation, timestamp);
+}
+
+// Creates a FrameBuffer from raw grayscale buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromGrayRawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation, const absl::Time timestamp) {
+  FrameBuffer::Plane input_plane = {/*buffer=*/input,
+                                    /*stride=*/{dimension.width, kGrayChannel}};
+  return FrameBuffer::Create({input_plane}, dimension,
+                             FrameBuffer::Format::kGRAY, orientation,
+                             timestamp);
+}
+
+// Creates a FrameBuffer from raw YUV buffer and passing arguments.
+StatusOr<std::unique_ptr<FrameBuffer>> CreateFromYuvRawBuffer(
+    const uint8* y_plane, const uint8* u_plane, const uint8* v_plane,
+    FrameBuffer::Format format, FrameBuffer::Dimension dimension,
+    int row_stride_y, int row_stride_uv, int pixel_stride_uv,
+    FrameBuffer::Orientation orientation, const absl::Time timestamp) {
+  const int pixel_stride_y = 1;
+  std::vector<FrameBuffer::Plane> planes;
+  if (format == FrameBuffer::Format::kNV21 ||
+      format == FrameBuffer::Format::kYV12) {
+    planes = {{y_plane, /*stride=*/{row_stride_y, pixel_stride_y}},
+              {v_plane, /*stride=*/{row_stride_uv, pixel_stride_uv}},
+              {u_plane, /*stride=*/{row_stride_uv, pixel_stride_uv}}};
+  } else if (format == FrameBuffer::Format::kNV12 ||
+             format == FrameBuffer::Format::kYV21) {
+    planes = {{y_plane, /*stride=*/{row_stride_y, pixel_stride_y}},
+              {u_plane, /*stride=*/{row_stride_uv, pixel_stride_uv}},
+              {v_plane, /*stride=*/{row_stride_uv, pixel_stride_uv}}};
+  } else {
+    return absl::InvalidArgumentError(
+        absl::StrFormat("Input format is not YUV-like: %i.", format));
+  }
+  return FrameBuffer::Create(planes, dimension, format, orientation, timestamp);
+}
+
+StatusOr<std::unique_ptr<FrameBuffer>> CreateFromRawBuffer(
+    const uint8* buffer, FrameBuffer::Dimension dimension,
+    const FrameBuffer::Format target_format,
+    FrameBuffer::Orientation orientation, absl::Time timestamp) {
+  switch (target_format) {
+    case FrameBuffer::Format::kNV12:
+      return CreateFromNV12RawBuffer(buffer, dimension, orientation, timestamp);
+    case FrameBuffer::Format::kNV21:
+      return CreateFromNV21RawBuffer(buffer, dimension, orientation, timestamp);
+    case FrameBuffer::Format::kYV12: {
+      ASSIGN_OR_RETURN(const FrameBuffer::Dimension uv_dimension,
+                       GetUvPlaneDimension(dimension, target_format));
+      return CreateFromYuvRawBuffer(
+          /*y_plane=*/buffer,
+          /*u_plane=*/buffer + dimension.Size() + uv_dimension.Size(),
+          /*v_plane=*/buffer + dimension.Size(), target_format, dimension,
+          /*row_stride_y=*/dimension.width, uv_dimension.width,
+          /*pixel_stride_uv=*/1, orientation, timestamp);
+    }
+    case FrameBuffer::Format::kYV21: {
+      ASSIGN_OR_RETURN(const FrameBuffer::Dimension uv_dimension,
+                       GetUvPlaneDimension(dimension, target_format));
+      return CreateFromYuvRawBuffer(
+          /*y_plane=*/buffer, /*u_plane=*/buffer + dimension.Size(),
+          /*v_plane=*/buffer + dimension.Size() + uv_dimension.Size(),
+          target_format, dimension, /*row_stride_y=*/dimension.width,
+          uv_dimension.width,
+          /*pixel_stride_uv=*/1, orientation, timestamp);
+    }
+    case FrameBuffer::Format::kRGBA:
+      return CreateFromRgbaRawBuffer(buffer, dimension, orientation, timestamp);
+    case FrameBuffer::Format::kRGB:
+      return CreateFromRgbRawBuffer(buffer, dimension, orientation, timestamp);
+    case FrameBuffer::Format::kGRAY:
+      return CreateFromGrayRawBuffer(buffer, dimension, orientation, timestamp);
+    default:
+
+      return absl::InternalError(
+          absl::StrFormat("Unsupported buffer format: %i.", target_format));
+  }
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h
new file mode 100644
index 0000000..e250d15
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h
@@ -0,0 +1,143 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_COMMON_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_COMMON_UTILS_H_
+
+#include <memory>
+
+#include "absl/status/status.h"
+#include "absl/time/clock.h"
+#include "absl/time/time.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+constexpr int kRgbaPixelBytes = 4, kRgbPixelBytes = 3, kGrayPixelBytes = 1;
+
+// Miscellaneous Methods
+// -----------------------------------------------------------------
+
+// Returns the frame buffer size in bytes based on the input format and
+// dimensions. GRAY, YV12/YV21 are in the planar formats, NV12/NV21 are in the
+// semi-planar formats with the interleaved UV planes. RGB/RGBA are in the
+// interleaved format.
+int GetFrameBufferByteSize(FrameBuffer::Dimension dimension,
+                           FrameBuffer::Format format);
+
+// Returns pixel stride info for kGRAY, kRGB, kRGBA formats.
+tflite::support::StatusOr<int> GetPixelStrides(FrameBuffer::Format format);
+
+// Returns the biplanar UV raw buffer for NV12/NV21 frame buffer.
+tflite::support::StatusOr<const uint8*> GetUvRawBuffer(
+    const FrameBuffer& buffer);
+
+// Returns U or V plane dimension with the given buffer `dimension` and
+// `format`. Only supports NV12/NV21/YV12/YV21 formats. Returns
+// InvalidArgumentError if 'dimension' is invalid or 'format' is other than the
+// supported formats. This method assums the UV plane share the same dimension,
+// especially for the YV12 / YV21 formats.
+tflite::support::StatusOr<FrameBuffer::Dimension> GetUvPlaneDimension(
+    FrameBuffer::Dimension dimension, FrameBuffer::Format format);
+
+// Returns crop dimension based on crop start and end points.
+FrameBuffer::Dimension GetCropDimension(int x0, int x1, int y0, int y1);
+
+// Validation Methods
+// -----------------------------------------------------------------
+
+// Validates that the given buffer has the correct metadata. Returns error
+// state when any buffer has missing stride info.
+absl::Status ValidateBufferPlaneMetadata(const FrameBuffer& buffer);
+
+// Validates that the given buffer has the correct format for its configuration.
+absl::Status ValidateBufferFormat(const FrameBuffer& buffer);
+
+// Validates that the given buffers have the correct format for their
+// configuration.
+absl::Status ValidateBufferFormats(const FrameBuffer& buffer1,
+                                   const FrameBuffer& buffer2);
+
+// Validates the given inputs for resizing `buffer`.
+absl::Status ValidateResizeBufferInputs(const FrameBuffer& buffer,
+                                        const FrameBuffer& output_buffer);
+
+// Validates the given inputs for rotating `buffer`.
+absl::Status ValidateRotateBufferInputs(const FrameBuffer& buffer,
+                                        const FrameBuffer& output_buffer,
+                                        int angle_deg);
+
+// Validates the given inputs for cropping `buffer`.
+//
+// (x0, y0) represents the top-left point of the buffer.
+// (x1, y1) represents the bottom-right point of the buffer.
+absl::Status ValidateCropBufferInputs(const FrameBuffer& buffer,
+                                      const FrameBuffer& output_buffer, int x0,
+                                      int y0, int x1, int y1);
+
+// Validates the given inputs for flipping `buffer` horizontally or vertically.
+absl::Status ValidateFlipBufferInputs(const FrameBuffer& buffer,
+                                      const FrameBuffer& output_buffer);
+
+// Validates that `from_format` can be converted to `to_format`.
+//
+// The given formats must not be equal.
+absl::Status ValidateConvertFormats(FrameBuffer::Format from_format,
+                                    FrameBuffer::Format to_format);
+
+// Creation Methods
+// -----------------------------------------------------------------
+
+// Creates a FrameBuffer from raw RGBA buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromRgbaRawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation = FrameBuffer::Orientation::kTopLeft,
+    absl::Time timestamp = absl::Now());
+
+// Creates a FrameBuffer from raw RGB buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromRgbRawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation = FrameBuffer::Orientation::kTopLeft,
+    absl::Time timestamp = absl::Now());
+
+// Creates a FrameBuffer from raw grayscale buffer and passing arguments.
+std::unique_ptr<FrameBuffer> CreateFromGrayRawBuffer(
+    const uint8* input, FrameBuffer::Dimension dimension,
+    FrameBuffer::Orientation orientation = FrameBuffer::Orientation::kTopLeft,
+    absl::Time timestamp = absl::Now());
+
+// Creates a FrameBuffer from raw YUV buffer and passing arguments.
+tflite::support::StatusOr<std::unique_ptr<FrameBuffer>> CreateFromYuvRawBuffer(
+    const uint8* y_plane, const uint8* u_plane, const uint8* v_plane,
+    FrameBuffer::Format format, FrameBuffer::Dimension dimension,
+    int row_stride_y, int row_stride_uv, int pixel_stride_uv,
+    FrameBuffer::Orientation orientation = FrameBuffer::Orientation::kTopLeft,
+    absl::Time timestamp = absl::Now());
+
+// Creates an instance of FrameBuffer from raw buffer and passing arguments.
+tflite::support::StatusOr<std::unique_ptr<FrameBuffer>> CreateFromRawBuffer(
+    const uint8* buffer, FrameBuffer::Dimension dimension,
+    FrameBuffer::Format target_format,
+    FrameBuffer::Orientation orientation = FrameBuffer::Orientation::kTopLeft,
+    absl::Time timestamp = absl::Now());
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_COMMON_UTILS_H_
diff --git a/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.cc b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.cc
new file mode 100644
index 0000000..9b9d830
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.cc
@@ -0,0 +1,619 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h"
+
+#include <algorithm>
+#include <iterator>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/memory/memory.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow/lite/kernels/op_macros.h"
+#include "tensorflow/lite/kernels/internal/compatibility.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+namespace {
+
+// Exif grouping to help determine rotation and flipping neededs between
+// different orientations.
+constexpr int kExifGroup[] = {1, 6, 3, 8, 2, 5, 4, 7};
+// Exif group size.
+constexpr int kExifGroupSize = 4;
+
+// Returns orientation position in Exif group.
+static int GetOrientationIndex(FrameBuffer::Orientation orientation) {
+  const int* index = std::find(kExifGroup, kExifGroup + kExifGroupSize * 2,
+                               static_cast<int>(orientation));
+  if (index < kExifGroup + kExifGroupSize * 2) {
+    return std::distance(kExifGroup, index);
+  }
+  return -1;
+}
+
+// Returns the coordinates of `box` respect to its containing image (dimension
+// defined by `width` and `height`) orientation change. The `angle` is defined
+// in counterclockwise degree in one of the values [0, 90, 180, 270].
+//
+// The below diagrams illustrate calling this method with 90 CCW degree.
+//
+// The [1]-[4] denotes image corners and 1 - 4 denotes the box corners. The *
+// denotes the current origin.
+//
+//             width
+//   [1]*----------------[2]
+//    |                   |
+//    |                   |
+//    |        1*-----2   | height
+//    |        | box  |   |
+//    |        3------4   |
+//   [3]-----------------[4]
+//
+// When rotate the above image by 90 CCW degree, the origin also changes
+// respects to its containing coordinate space.
+//
+//         height
+//   [2]*----------[4]
+//    |             |
+//    |     2*---4  |
+//    |     |box |  |
+//    |     |    |  | width
+//    |     1----3  |
+//    |             |
+//    |             |
+//    |             |
+//   [1]-----------[3]
+//
+// The origin is always defined by the top left corner. After rotation, the
+// box origin changed from 1 to 2.
+// The new box origin is (x:box.origin_y, y:width - (box.origin_x + box.width).
+// The new box dimension is (w: box.height, h: box.width).
+//
+static BoundingBox RotateBoundingBox(const BoundingBox& box, int angle,
+                                     FrameBuffer::Dimension frame_dimension) {
+  int rx = box.origin_x(), ry = box.origin_y(), rw = box.width(),
+      rh = box.height();
+  const int box_right_bound =
+      frame_dimension.width - (box.origin_x() + box.width());
+  const int box_bottom_bound =
+      frame_dimension.height - (box.origin_y() + box.height());
+  switch (angle) {
+    case 90:
+      rx = box.origin_y();
+      ry = box_right_bound;
+      using std::swap;
+      swap(rw, rh);
+      break;
+    case 180:
+      rx = box_right_bound;
+      ry = box_bottom_bound;
+      break;
+    case 270:
+      rx = box_bottom_bound;
+      ry = box.origin_x();
+      using std::swap;
+      swap(rw, rh);
+      break;
+  }
+  BoundingBox result;
+  result.set_origin_x(rx);
+  result.set_origin_y(ry);
+  result.set_width(rw);
+  result.set_height(rh);
+  return result;
+}
+
+// Returns the input coordinates with respect to its containing image (dimension
+// defined by `width` and `height`) orientation change. The `angle` is defined
+// in counterclockwise degree in one of the values [0, 90, 180, 270].
+//
+// See `RotateBoundingBox` above for more details.
+static void RotateCoordinates(int from_x, int from_y, int angle,
+                              const FrameBuffer::Dimension& frame_dimension,
+                              int* to_x, int* to_y) {
+  switch (angle) {
+    case 0:
+      *to_x = from_x;
+      *to_y = from_y;
+      break;
+    case 90:
+      *to_x = from_y;
+      *to_y = frame_dimension.width - from_x - 1;
+      break;
+    case 180:
+      *to_x = frame_dimension.width - from_x - 1;
+      *to_y = frame_dimension.height - from_y - 1;
+      break;
+    case 270:
+      *to_x = frame_dimension.height - from_y - 1;
+      *to_y = from_x;
+      break;
+  }
+}
+
+}  // namespace
+
+int GetBufferByteSize(FrameBuffer::Dimension dimension,
+                      FrameBuffer::Format format) {
+  return GetFrameBufferByteSize(dimension, format);
+}
+
+FrameBufferUtils::FrameBufferUtils(ProcessEngine engine) {
+  switch (engine) {
+    case ProcessEngine::kLibyuv:
+      utils_ = absl::make_unique<LibyuvFrameBufferUtils>();
+      break;
+    default:
+      TF_LITE_FATAL(
+          absl::StrFormat("Unexpected ProcessEngine: %d.", engine).c_str());
+  }
+}
+
+BoundingBox OrientBoundingBox(const BoundingBox& from_box,
+                              FrameBuffer::Orientation from_orientation,
+                              FrameBuffer::Orientation to_orientation,
+                              FrameBuffer::Dimension from_dimension) {
+  BoundingBox to_box = from_box;
+  OrientParams params = GetOrientParams(from_orientation, to_orientation);
+  // First, rotate if needed.
+  if (params.rotation_angle_deg > 0) {
+    to_box =
+        RotateBoundingBox(to_box, params.rotation_angle_deg, from_dimension);
+  }
+  // Then perform horizontal or vertical flip if needed.
+  FrameBuffer::Dimension to_dimension = from_dimension;
+  if (params.rotation_angle_deg == 90 || params.rotation_angle_deg == 270) {
+    to_dimension.Swap();
+  }
+  if (params.flip == OrientParams::FlipType::kVertical) {
+    to_box.set_origin_y(to_dimension.height -
+                        (to_box.origin_y() + to_box.height()));
+  }
+  if (params.flip == OrientParams::FlipType::kHorizontal) {
+    to_box.set_origin_x(to_dimension.width -
+                        (to_box.origin_x() + to_box.width()));
+  }
+  return to_box;
+}
+
+BoundingBox OrientAndDenormalizeBoundingBox(
+    float from_left, float from_top, float from_right, float from_bottom,
+    FrameBuffer::Orientation from_orientation,
+    FrameBuffer::Orientation to_orientation,
+    FrameBuffer::Dimension from_dimension) {
+  BoundingBox from_box;
+  from_box.set_origin_x(from_left * from_dimension.width);
+  from_box.set_origin_y(from_top * from_dimension.height);
+  from_box.set_width(round(abs(from_right - from_left) * from_dimension.width));
+  from_box.set_height(
+      round(abs(from_bottom - from_top) * from_dimension.height));
+  BoundingBox to_box = OrientBoundingBox(from_box, from_orientation,
+                                         to_orientation, from_dimension);
+  return to_box;
+}
+
+void OrientCoordinates(int from_x, int from_y,
+                       FrameBuffer::Orientation from_orientation,
+                       FrameBuffer::Orientation to_orientation,
+                       FrameBuffer::Dimension from_dimension, int* to_x,
+                       int* to_y) {
+  *to_x = from_x;
+  *to_y = from_y;
+  OrientParams params = GetOrientParams(from_orientation, to_orientation);
+  // First, rotate if needed.
+  if (params.rotation_angle_deg > 0) {
+    RotateCoordinates(from_x, from_y, params.rotation_angle_deg, from_dimension,
+                      to_x, to_y);
+  }
+  // Then perform horizontal or vertical flip if needed.
+  FrameBuffer::Dimension to_dimension = from_dimension;
+  if (params.rotation_angle_deg == 90 || params.rotation_angle_deg == 270) {
+    to_dimension.Swap();
+  }
+  if (params.flip == OrientParams::FlipType::kVertical) {
+    *to_y = to_dimension.height - *to_y - 1;
+  }
+  if (params.flip == OrientParams::FlipType::kHorizontal) {
+    *to_x = to_dimension.width - *to_x - 1;
+  }
+}
+
+// The algorithm is based on grouping orientations into two groups with specific
+// order. The two groups of orientation are {1, 6, 3, 8} and {2, 5, 4, 7}. See
+// image (https://www.impulseadventure.com/photo/images/orient_flag.gif) for
+// the visual grouping illustration.
+//
+// Each group contains elements can be transformed into one another by rotation.
+// The elements order within a group is important such that the distance between
+// the elements indicates the multiples of 90 degree needed to orient from one
+// element to another. For example, to orient element 1 to element 6, a 90
+// degree CCW rotation is needed.
+//
+// The corresponding order between the two groups is important such that the
+// even index defined the need for horizontal flipping and the odd index defined
+// the need for vertical flipping. For example, to orient element 1 to element 2
+// (even index) a horizontal flipping is needed.
+//
+// The implementation determines the group and element index of from and to
+// orientations. Based on the group and element index information, the above
+// characteristic is used to calculate the rotation angle and the need for
+// horizontal or vertical flipping.
+OrientParams GetOrientParams(FrameBuffer::Orientation from_orientation,
+                             FrameBuffer::Orientation to_orientation) {
+  int from_index = GetOrientationIndex(from_orientation);
+  int to_index = GetOrientationIndex(to_orientation);
+  int angle = 0;
+  absl::optional<OrientParams::FlipType> flip;
+
+  TFLITE_DCHECK(from_index > -1 && to_index > -1);
+
+  if ((from_index < kExifGroupSize && to_index < kExifGroupSize) ||
+      (from_index >= kExifGroupSize && to_index >= kExifGroupSize)) {
+    // Only needs rotation.
+
+    // The orientations' position differences translates to how many
+    // multiple of 90 degrees it needs for conversion. The position difference
+    // calculation within a group is circular.
+    angle = (kExifGroupSize - (from_index - to_index)) % kExifGroupSize * 90;
+  } else {
+    // Needs rotation and flipping.
+    int from_index_mod = from_index % kExifGroupSize;
+    int to_index_mod = to_index % kExifGroupSize;
+    angle = (kExifGroupSize - (from_index_mod - to_index_mod)) %
+            kExifGroupSize * 90;
+    if (to_index_mod % 2 == 1) {
+      flip = OrientParams::FlipType::kVertical;
+    } else {
+      flip = OrientParams::FlipType::kHorizontal;
+    }
+  }
+  return {angle, flip};
+}
+
+bool RequireDimensionSwap(FrameBuffer::Orientation from_orientation,
+                          FrameBuffer::Orientation to_orientation) {
+  OrientParams params = GetOrientParams(from_orientation, to_orientation);
+  return params.rotation_angle_deg == 90 || params.rotation_angle_deg == 270;
+}
+
+absl::Status FrameBufferUtils::Crop(const FrameBuffer& buffer, int x0, int y0,
+                                    int x1, int y1,
+                                    FrameBuffer* output_buffer) {
+  TFLITE_DCHECK(utils_ != nullptr);
+  return utils_->Crop(buffer, x0, y0, x1, y1, output_buffer);
+}
+
+FrameBuffer::Dimension FrameBufferUtils::GetSize(
+    const FrameBuffer& buffer, const FrameBufferOperation& operation) {
+  FrameBuffer::Dimension dimension = buffer.dimension();
+  if (absl::holds_alternative<OrientOperation>(operation)) {
+    OrientParams params =
+        GetOrientParams(buffer.orientation(),
+                        absl::get<OrientOperation>(operation).to_orientation);
+    if (params.rotation_angle_deg == 90 || params.rotation_angle_deg == 270) {
+      dimension.Swap();
+    }
+  } else if (absl::holds_alternative<CropResizeOperation>(operation)) {
+    const auto& crop_resize = absl::get<CropResizeOperation>(operation);
+    dimension = crop_resize.resize_dimension;
+  }
+  return dimension;
+}
+
+std::vector<FrameBuffer::Plane> FrameBufferUtils::GetPlanes(
+    const uint8* buffer, FrameBuffer::Dimension dimension,
+    FrameBuffer::Format format) {
+  std::vector<FrameBuffer::Plane> planes;
+  switch (format) {
+    case FrameBuffer::Format::kGRAY:
+      planes.push_back({/*buffer=*/buffer,
+                        /*stride=*/{/*row_stride_bytes=*/dimension.width * 1,
+                                    /*pixel_stride_bytes=*/1}});
+      break;
+    case FrameBuffer::Format::kRGB:
+      planes.push_back({/*buffer=*/buffer,
+                        /*stride=*/{/*row_stride_bytes=*/dimension.width * 3,
+                                    /*pixel_stride_bytes=*/3}});
+      break;
+    case FrameBuffer::Format::kRGBA:
+      planes.push_back({/*buffer=*/buffer,
+                        /*stride=*/{/*row_stride_bytes=*/dimension.width * 4,
+                                    /*pixel_stride_bytes=*/4}});
+      break;
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kNV12: {
+      planes.push_back(
+          {buffer, /*stride=*/{/*row_stride_bytes=*/dimension.width,
+                               /*pixel_stride_bytes=*/1}});
+      planes.push_back({buffer + (dimension.width * dimension.height),
+                        /*stride=*/{/*row_stride_bytes=*/dimension.width,
+                                    /*pixel_stride_bytes=*/2}});
+    } break;
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21: {
+      const int y_buffer_size = dimension.width * dimension.height;
+      const int uv_row_stride = (dimension.width + 1) / 2;
+      const int uv_buffer_size = uv_row_stride * (dimension.height + 1) / 2;
+      planes.push_back(
+          {buffer, /*stride=*/{/*row_stride_bytes=*/dimension.width,
+                               /*pixel_stride_bytes=*/1}});
+      planes.push_back(
+          {buffer + y_buffer_size, /*stride=*/{
+               /*row_stride_bytes=*/uv_row_stride, /*pixel_stride_bytes=*/1}});
+      planes.push_back(
+          {buffer + y_buffer_size + uv_buffer_size, /*stride=*/{
+               /*row_stride_bytes=*/uv_row_stride, /*pixel_stride_bytes=*/1}});
+    } break;
+    default:
+      break;
+  }
+  return planes;
+}
+
+FrameBuffer::Orientation FrameBufferUtils::GetOrientation(
+    const FrameBuffer& buffer, const FrameBufferOperation& operation) {
+  if (absl::holds_alternative<OrientOperation>(operation)) {
+    return absl::get<OrientOperation>(operation).to_orientation;
+  }
+  return buffer.orientation();
+}
+
+FrameBuffer::Format FrameBufferUtils::GetFormat(
+    const FrameBuffer& buffer, const FrameBufferOperation& operation) {
+  if (absl::holds_alternative<ConvertOperation>(operation)) {
+    return absl::get<ConvertOperation>(operation).to_format;
+  }
+  return buffer.format();
+}
+
+absl::Status FrameBufferUtils::Execute(const FrameBuffer& buffer,
+                                       const FrameBufferOperation& operation,
+                                       FrameBuffer* output_buffer) {
+  if (absl::holds_alternative<CropResizeOperation>(operation)) {
+    const auto& params = absl::get<CropResizeOperation>(operation);
+    RETURN_IF_ERROR(
+        Crop(buffer, params.crop_origin_x, params.crop_origin_y,
+             (params.crop_dimension.width + params.crop_origin_x - 1),
+             (params.crop_dimension.height + params.crop_origin_y - 1),
+             output_buffer));
+  } else if (absl::holds_alternative<ConvertOperation>(operation)) {
+    RETURN_IF_ERROR(Convert(buffer, output_buffer));
+  } else if (absl::holds_alternative<OrientOperation>(operation)) {
+    RETURN_IF_ERROR(Orient(buffer, output_buffer));
+  } else {
+    return absl::UnimplementedError(absl::StrFormat(
+        "FrameBufferOperation %i is not supported.", operation.index()));
+  }
+  return absl::OkStatus();
+}
+
+absl::Status FrameBufferUtils::Resize(const FrameBuffer& buffer,
+                                      FrameBuffer* output_buffer) {
+  TFLITE_DCHECK(utils_ != nullptr);
+  return utils_->Resize(buffer, output_buffer);
+}
+
+absl::Status FrameBufferUtils::Rotate(const FrameBuffer& buffer,
+                                      RotationDegree rotation,
+                                      FrameBuffer* output_buffer) {
+  TFLITE_DCHECK(utils_ != nullptr);
+  return utils_->Rotate(buffer, 90 * static_cast<int>(rotation), output_buffer);
+}
+
+absl::Status FrameBufferUtils::FlipHorizontally(const FrameBuffer& buffer,
+                                                FrameBuffer* output_buffer) {
+  TFLITE_DCHECK(utils_ != nullptr);
+  return utils_->FlipHorizontally(buffer, output_buffer);
+}
+
+absl::Status FrameBufferUtils::FlipVertically(const FrameBuffer& buffer,
+                                              FrameBuffer* output_buffer) {
+  TFLITE_DCHECK(utils_ != nullptr);
+  return utils_->FlipVertically(buffer, output_buffer);
+}
+
+absl::Status FrameBufferUtils::Convert(const FrameBuffer& buffer,
+                                       FrameBuffer* output_buffer) {
+  TFLITE_DCHECK(utils_ != nullptr);
+  return utils_->Convert(buffer, output_buffer);
+}
+
+absl::Status FrameBufferUtils::Orient(const FrameBuffer& buffer,
+                                      FrameBuffer* output_buffer) {
+  TFLITE_DCHECK(utils_ != nullptr);
+
+  OrientParams params =
+      GetOrientParams(buffer.orientation(), output_buffer->orientation());
+  if (params.rotation_angle_deg == 0 && !params.flip.has_value()) {
+    // If no rotation or flip is needed, we will copy the buffer to
+    // output_buffer.
+    return utils_->Resize(buffer, output_buffer);
+  }
+
+  if (params.rotation_angle_deg == 0) {
+    // Only perform flip operation.
+    switch (*params.flip) {
+      case OrientParams::FlipType::kHorizontal:
+        return utils_->FlipHorizontally(buffer, output_buffer);
+      case OrientParams::FlipType::kVertical:
+        return utils_->FlipVertically(buffer, output_buffer);
+    }
+  }
+
+  if (!params.flip.has_value()) {
+    // Only perform rotation operation.
+    return utils_->Rotate(buffer, params.rotation_angle_deg, output_buffer);
+  }
+
+  // Perform rotation and flip operations.
+  // Create a temporary buffer to hold the rotation result.
+  auto tmp_buffer = absl::make_unique<uint8[]>(
+      GetBufferByteSize(output_buffer->dimension(), output_buffer->format()));
+  auto tmp_frame_buffer = FrameBuffer::Create(
+      GetPlanes(tmp_buffer.get(), output_buffer->dimension(),
+                output_buffer->format()),
+      output_buffer->dimension(), buffer.format(), buffer.orientation());
+
+  RETURN_IF_ERROR(utils_->Rotate(buffer, params.rotation_angle_deg,
+                                 tmp_frame_buffer.get()));
+  if (params.flip == OrientParams::FlipType::kHorizontal) {
+    return utils_->FlipHorizontally(*tmp_frame_buffer, output_buffer);
+  } else {
+    return utils_->FlipVertically(*tmp_frame_buffer, output_buffer);
+  }
+}
+
+absl::Status FrameBufferUtils::Execute(
+    const FrameBuffer& buffer,
+    const std::vector<FrameBufferOperation>& operations,
+    FrameBuffer* output_buffer) {
+  // Reference variables to swapping input and output buffers for each command.
+  FrameBuffer input_frame_buffer = buffer;
+  FrameBuffer temp_frame_buffer = buffer;
+
+  // Temporary buffers and its size to hold intermediate results.
+  int buffer1_size = 0;
+  int buffer2_size = 0;
+  std::unique_ptr<uint8[]> buffer1;
+  std::unique_ptr<uint8[]> buffer2;
+
+  for (int i = 0; i < operations.size(); i++) {
+    const FrameBufferOperation& operation = operations[i];
+
+    // The first command's input is always passed in `buffer`. Before
+    // process each command, the input_frame_buffer is pointed at the previous
+    // command's output buffer.
+    if (i == 0) {
+      input_frame_buffer = buffer;
+    } else {
+      input_frame_buffer = temp_frame_buffer;
+    }
+
+    // Calculates the resulting metadata from the command and the input.
+    FrameBuffer::Dimension new_size = GetSize(input_frame_buffer, operation);
+    FrameBuffer::Orientation new_orientation =
+        GetOrientation(input_frame_buffer, operation);
+    FrameBuffer::Format new_format = GetFormat(input_frame_buffer, operation);
+    int byte_size = GetBufferByteSize(new_size, new_format);
+
+    // The last command's output buffer is always passed in `output_buffer`.
+    // For other commands, we create temporary FrameBuffer for processing.
+    if ((i + 1) == operations.size()) {
+      temp_frame_buffer = *output_buffer;
+      // Validate the `output_buffer` metadata mathes with command line chain
+      // resulting metadata.
+      if (temp_frame_buffer.format() != new_format ||
+          temp_frame_buffer.orientation() != new_orientation ||
+          temp_frame_buffer.dimension() != new_size) {
+        return absl::InvalidArgumentError(
+            "The output metadata does not match pipeline result metadata.");
+      }
+    } else {
+      // Create a temporary buffer to hold intermediate results. For simplicity,
+      // we only create one continuous memory with no padding for intermediate
+      // results.
+      //
+      // We hold maximum 2 temporary buffers in memory at any given time.
+      //
+      // The pipeline is a linear chain. The output buffer from previous command
+      // becomes the input buffer for the next command. We simply use odd / even
+      // index to swap between buffers.
+      std::vector<FrameBuffer::Plane> planes;
+      if (i % 2 == 0) {
+        if (buffer1_size < byte_size) {
+          buffer1_size = byte_size;
+          buffer1 = absl::make_unique<uint8[]>(byte_size);
+        }
+        planes = GetPlanes(buffer1.get(), new_size, new_format);
+      } else {
+        if (buffer2_size < byte_size) {
+          buffer2_size = byte_size;
+          buffer2 = absl::make_unique<uint8[]>(byte_size);
+        }
+        planes = GetPlanes(buffer2.get(), new_size, new_format);
+      }
+      if (planes.empty()) {
+        return absl::InternalError("Failed to construct temporary buffer.");
+      }
+      temp_frame_buffer = FrameBuffer(planes, new_size, new_format,
+                                      new_orientation, buffer.timestamp());
+    }
+    RETURN_IF_ERROR(Execute(input_frame_buffer, operation, &temp_frame_buffer));
+  }
+  return absl::OkStatus();
+}
+
+absl::Status FrameBufferUtils::Preprocess(
+    const FrameBuffer& buffer, absl::optional<BoundingBox> bounding_box,
+    FrameBuffer* output_buffer) {
+  std::vector<FrameBufferOperation> frame_buffer_operations;
+  // Handle cropping and resizing.
+  bool needs_dimension_swap =
+      RequireDimensionSwap(buffer.orientation(), output_buffer->orientation());
+  // For intermediate steps, we need to use dimensions based on the input
+  // orientation.
+  FrameBuffer::Dimension pre_orient_dimension = output_buffer->dimension();
+  if (needs_dimension_swap) {
+    pre_orient_dimension.Swap();
+  }
+
+  if (bounding_box.has_value()) {
+    // Cropping case.
+    frame_buffer_operations.push_back(CropResizeOperation(
+        bounding_box.value().origin_x(), bounding_box.value().origin_y(),
+        FrameBuffer::Dimension{bounding_box.value().width(),
+                               bounding_box.value().height()},
+        pre_orient_dimension));
+  } else if (pre_orient_dimension != buffer.dimension()) {
+    // Resizing case.
+    frame_buffer_operations.push_back(
+        CropResizeOperation(0, 0, buffer.dimension(), pre_orient_dimension));
+  }
+
+  // Handle color space conversion.
+  if (output_buffer->format() != buffer.format()) {
+    frame_buffer_operations.push_back(
+        ConvertOperation(output_buffer->format()));
+  }
+
+  // Handle orientation conversion.
+  if (output_buffer->orientation() != buffer.orientation()) {
+    frame_buffer_operations.push_back(
+        OrientOperation(output_buffer->orientation()));
+  }
+
+  // Execute the processing pipeline.
+  if (frame_buffer_operations.empty()) {
+    // Using resize to perform copy.
+    RETURN_IF_ERROR(Resize(buffer, output_buffer));
+  } else {
+    RETURN_IF_ERROR(Execute(buffer, frame_buffer_operations, output_buffer));
+  }
+  return absl::OkStatus();
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h
new file mode 100644
index 0000000..90a7491
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils.h
@@ -0,0 +1,292 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_UTILS_H_
+
+#include <memory>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/types/optional.h"
+#include "absl/types/variant.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils_interface.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Returns the minimal buffer size for a plane in bytes based on the given
+// format and dimensions.
+int GetBufferByteSize(FrameBuffer::Dimension dimension,
+                      FrameBuffer::Format format);
+
+// Rotates the `from_box` in `from_orientation` to `to_orientation` within an
+// image of size `from_dimension`.
+BoundingBox OrientBoundingBox(const BoundingBox& from_box,
+                              FrameBuffer::Orientation from_orientation,
+                              FrameBuffer::Orientation to_orientation,
+                              FrameBuffer::Dimension from_dimension);
+
+// Same as OrientBoundingBox but from normalized coordinates.
+BoundingBox OrientAndDenormalizeBoundingBox(
+    float from_left, float from_top, float from_right, float from_bottom,
+    FrameBuffer::Orientation from_orientation,
+    FrameBuffer::Orientation to_orientation,
+    FrameBuffer::Dimension from_dimension);
+
+// Rotates `(from_x, from_y)` coordinates from an image of dimension
+// `from_dimension` and orientation `from_orientation` into `(to_x, to_y)`
+// coordinates with orientation `to_orientation`.
+void OrientCoordinates(int from_x, int from_y,
+                       FrameBuffer::Orientation from_orientation,
+                       FrameBuffer::Orientation to_orientation,
+                       FrameBuffer::Dimension from_dimension, int* to_x,
+                       int* to_y);
+
+// Returns whether the conversion from from_orientation to to_orientation
+// requires 90 or 270 degrees rotation.
+bool RequireDimensionSwap(FrameBuffer::Orientation from_orientation,
+                          FrameBuffer::Orientation to_orientation);
+
+// Structure to express parameters needed to achieve orientation conversion.
+struct OrientParams {
+  // Counterclockwise rotation angle in degrees. This is expressed as a
+  // multiple of 90 degrees.
+  int rotation_angle_deg;
+  // Flipping operation. It must come after the rotation.
+  enum class FlipType { kHorizontal, kVertical };
+  absl::optional<FlipType> flip;
+};
+
+// Returns rotation angle and the need for horizontal flipping or vertical
+// flipping.
+OrientParams GetOrientParams(FrameBuffer::Orientation from_orientation,
+                             FrameBuffer::Orientation to_orientation);
+
+// The parameters needed to crop / resize.
+//
+// The coordinate system has its origin at the upper left corner, and
+// positive values extend down and to the right from it.
+//
+// After the operation, the `crop_origin` will become the new origin.
+// `crop_width` and `crop_height` defines the desired cropping region. After
+// cropping, a resize is performed based on the `resize_width` and
+// `resize_height`.
+//
+// To perform just cropping, the `crop_width` and `crop_height` should be the
+// same as `resize_width` `and resize_height`.
+struct CropResizeOperation {
+  CropResizeOperation(int crop_origin_x, int crop_origin_y,
+                      FrameBuffer::Dimension crop_dimension,
+                      FrameBuffer::Dimension resize_dimension)
+      : crop_origin_x(crop_origin_x),
+        crop_origin_y(crop_origin_y),
+        crop_dimension(crop_dimension),
+        resize_dimension(resize_dimension) {}
+
+  int crop_origin_x;
+  int crop_origin_y;
+  FrameBuffer::Dimension crop_dimension;
+  FrameBuffer::Dimension resize_dimension;
+};
+
+// The parameters needed to convert to the specified format.
+struct ConvertOperation {
+  explicit ConvertOperation(FrameBuffer::Format to_format)
+      : to_format(to_format) {}
+  FrameBuffer::Format to_format;
+};
+
+// The parameters needed to change the orientation.
+struct OrientOperation {
+  explicit OrientOperation(FrameBuffer::Orientation to_orientation)
+      : to_orientation(to_orientation) {}
+  FrameBuffer::Orientation to_orientation;
+};
+
+// A variant of the supported operations on FrameBuffers. Alias for user
+// convenience.
+using FrameBufferOperation =
+    absl::variant<CropResizeOperation, ConvertOperation, OrientOperation>;
+
+// Image processing utility. This utility provides both basic image buffer
+// manipulations (e.g. rotation, format conversion, resizing, etc) as well as
+// capability for chaining pipeline executions. The actual buffer processing
+// engine is configurable to allow optimization based on platforms.
+//
+// Examples:
+//
+//  // Create an instance of FrameBufferUtils with Halide processing engine.
+//  std::unique_ptr<FrameBufferUtils> utils = FrameBufferUtils::Create(kHalide);
+//
+//  // Perform single basic operation by each individual call.
+//  std::unique_ptr<FrameBuffer> input = FrameBuffer::Create(...);
+//  std::unique_ptr<FrameBuffer> output = FrameBuffer::Create(...);
+//  utils->Orient(*input, output.get());
+//  utils->Resize(*input, output.get());
+//
+//  // Chaining processing operations.
+//  const std::vector<FrameBufferOperation> operations = {
+//      ConvertOperation(FrameBuffer::Format::kNV21),
+//      CropResizeOperation(/*crop_origin_x=*/20, /*crop_origin_y=*/20,
+//                          /*crop_width=*/10, /*crop_height=*/10,
+//                          /*resize_width=*/10, /*resize_height=*/10),
+//      OrientOperation(FrameBuffer::Orientation::kLeftTop)};
+//  utils->Execute(*input, operations, output.get());
+class FrameBufferUtils {
+ public:
+  // Counter-clockwise rotation in degree.
+  enum class RotationDegree { k0 = 0, k90 = 1, k180 = 2, k270 = 3 };
+
+  // Underlying process engine used for performing operations.
+  enum class ProcessEngine {
+    kLibyuv,
+  };
+
+  // Factory method FrameBufferUtils instance. The processing engine is
+  // defined by `engine`.
+  static std::unique_ptr<FrameBufferUtils> Create(ProcessEngine engine) {
+    return absl::make_unique<FrameBufferUtils>(engine);
+  }
+
+  explicit FrameBufferUtils(ProcessEngine engine);
+
+  // Performs cropping operation.
+  //
+  // The coordinate system has its origin at the upper left corner, and
+  // positive values extend down and to the right from it. After cropping,
+  // (x0, y0) becomes (0, 0). The new width and height are
+  // (x1 - x0 + 1, y1 - y0 + 1).
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result. If the `output_buffer`
+  // size dimension does not match with crop dimension, then a resize is
+  // automatically performed.
+  absl::Status Crop(const FrameBuffer& buffer, int x0, int y0, int x1, int y1,
+                    FrameBuffer* output_buffer);
+
+  // Performs resizing operation.
+  //
+  // The resize dimension is determined based on output_buffer's size metadata.
+  //
+  // The output_buffer should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  absl::Status Resize(const FrameBuffer& buffer, FrameBuffer* output_buffer);
+
+  // Performs rotation operation.
+  //
+  // The rotation is specified in counter-clockwise direction.
+  //
+  // The output_buffer should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  absl::Status Rotate(const FrameBuffer& buffer, RotationDegree rotation,
+                      FrameBuffer* output_buffer);
+
+  // Performs horizontal flip operation.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  absl::Status FlipHorizontally(const FrameBuffer& buffer,
+                                FrameBuffer* output_buffer);
+
+  // Performs vertical flip operation.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  absl::Status FlipVertically(const FrameBuffer& buffer,
+                              FrameBuffer* output_buffer);
+
+  // Performs buffer format conversion.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  absl::Status Convert(const FrameBuffer& buffer, FrameBuffer* output_buffer);
+
+  // Performs buffer orientation conversion. Depends on the orientations, this
+  // method may perform rotation and optional flipping operations.
+  //
+  // If `buffer` and `output_buffer` has the same orientation, then a copy
+  // operation will performed.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  absl::Status Orient(const FrameBuffer& buffer, FrameBuffer* output_buffer);
+
+  // Performs the image processing operations specified, in that order.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  absl::Status Execute(const FrameBuffer& buffer,
+                       const std::vector<FrameBufferOperation>& operations,
+                       FrameBuffer* output_buffer);
+
+  // Performs a chain of operations to convert `buffer` to desired metadata
+  // (width, height, format, orientation) defined by `output_buffer` and
+  // optional cropping (`bounding_box`).
+  //
+  // Internally, a chain of operations is constructed. For performance
+  // optimization, operations are performed in the following order: crop,
+  // resize, convert color space format, and rotate.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result. Insufficient backing
+  // buffer size may cause garbage result or crash. Use `GetBufferByteSize` to
+  // calculate the minimal buffer size.
+  //
+  // If the `buffer` is already in desired format, then an extra copy will be
+  // performed.
+  //
+  // The input param `bounding_box` is defined in the `buffer` coordinate space.
+  absl::Status Preprocess(const FrameBuffer& buffer,
+                          absl::optional<BoundingBox> bounding_box,
+                          FrameBuffer* output_buffer);
+
+ private:
+  // Returns the new FrameBuffer size after the operation is applied.
+  FrameBuffer::Dimension GetSize(const FrameBuffer& buffer,
+                                 const FrameBufferOperation& operation);
+
+  // Returns the new FrameBuffer orientation after command is processed.
+  FrameBuffer::Orientation GetOrientation(
+      const FrameBuffer& buffer, const FrameBufferOperation& operation);
+
+  // Returns the new FrameBuffer format after command is processed.
+  FrameBuffer::Format GetFormat(const FrameBuffer& buffer,
+                                const FrameBufferOperation& operation);
+
+  // Returns Plane struct based on one dimension buffer and its metadata. If
+  // an error occurred, it will return an empty vector.
+  std::vector<FrameBuffer::Plane> GetPlanes(const uint8* buffer,
+                                            FrameBuffer::Dimension dimension,
+                                            FrameBuffer::Format format);
+
+  // Executes command with params.
+  absl::Status Execute(const FrameBuffer& buffer,
+                       const FrameBufferOperation& operation,
+                       FrameBuffer* output_buffer);
+
+  // Execution engine conforms to FrameBufferUtilsInterface.
+  std::unique_ptr<FrameBufferUtilsInterface> utils_;
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_UTILS_H_
diff --git a/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils_interface.h b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils_interface.h
new file mode 100644
index 0000000..502e998
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils_interface.h
@@ -0,0 +1,88 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_UTILS_INTERFACE_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_UTILS_INTERFACE_H_
+
+#include "absl/status/status.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Interface for the FrameBuffer image processing library.
+class FrameBufferUtilsInterface {
+ public:
+  virtual ~FrameBufferUtilsInterface() = default;
+
+  // Crops `buffer` to the specified points.
+  //
+  // The coordinate system has its origin at the upper left corner, and
+  // positive values extend down and to the right from it. After cropping,
+  // the top left point becomes (0, 0). The new width and height are
+  // (x1 - x0 + 1, y1 - y0 + 1).
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  virtual absl::Status Crop(const FrameBuffer& buffer, int x0, int y0, int x1,
+                            int y1, FrameBuffer* output_buffer) = 0;
+
+  // Resizes `buffer` to the size of the given `output_buffer`.
+  //
+  // The resize dimension is determined based on the size of `output_buffer`.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  virtual absl::Status Resize(const FrameBuffer& buffer,
+                              FrameBuffer* output_buffer) = 0;
+
+  // Rotates `buffer` counter-clockwise by the given `angle_deg` (in degrees).
+  //
+  // When rotating by 90 degrees, the top-right corner of `buffer` becomes
+  // the top-left corner of `output_buffer`. The given angle must be a multiple
+  // of 90 degrees.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  virtual absl::Status Rotate(const FrameBuffer& buffer, int angle_deg,
+                              FrameBuffer* output_buffer) = 0;
+
+  // Flips `buffer` horizontally.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  virtual absl::Status FlipHorizontally(const FrameBuffer& buffer,
+                                        FrameBuffer* output_buffer) = 0;
+
+  // Flips `buffer` vertically.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  virtual absl::Status FlipVertically(const FrameBuffer& buffer,
+                                      FrameBuffer* output_buffer) = 0;
+
+  // Converts `buffer`'s format to the format of the given `output_buffer`.
+  //
+  // The `output_buffer` should have metadata populated and its backing buffer
+  // should be big enough to store the operation result.
+  virtual absl::Status Convert(const FrameBuffer& buffer,
+                               FrameBuffer* output_buffer) = 0;
+};
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_FRAME_BUFFER_UTILS_INTERFACE_H_
diff --git a/tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.cc b/tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.cc
new file mode 100644
index 0000000..51e72fa
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.cc
@@ -0,0 +1,254 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.h"
+
+#include "absl/status/status.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+namespace {
+
+using ::absl::StatusCode;
+using ::tflite::ColorSpaceType_RGB;
+using ::tflite::ContentProperties;
+using ::tflite::ContentProperties_ImageProperties;
+using ::tflite::EnumNameContentProperties;
+using ::tflite::ImageProperties;
+using ::tflite::TensorMetadata;
+using ::tflite::metadata::ModelMetadataExtractor;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+using ::tflite::task::core::TfLiteEngine;
+
+StatusOr<const TensorMetadata*> GetInputTensorMetadataIfAny(
+    const ModelMetadataExtractor& metadata_extractor) {
+  if (metadata_extractor.GetModelMetadata() == nullptr ||
+      metadata_extractor.GetModelMetadata()->subgraph_metadata() == nullptr) {
+    // Some models have no metadata at all (or very partial), so exit early.
+    return nullptr;
+  } else if (metadata_extractor.GetInputTensorCount() != 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Models are assumed to have a single input TensorMetadata.",
+        TfLiteSupportStatus::kInvalidNumInputTensorsError);
+  }
+
+  const TensorMetadata* metadata = metadata_extractor.GetInputTensorMetadata(0);
+
+  if (metadata == nullptr) {
+    // Should never happen.
+    return CreateStatusWithPayload(StatusCode::kInternal,
+                                   "Input TensorMetadata is null.");
+  }
+
+  return metadata;
+}
+
+StatusOr<const ImageProperties*> GetImagePropertiesIfAny(
+    const TensorMetadata& tensor_metadata) {
+  if (tensor_metadata.content() == nullptr ||
+      tensor_metadata.content()->content_properties() == nullptr) {
+    return nullptr;
+  }
+
+  ContentProperties type = tensor_metadata.content()->content_properties_type();
+
+  if (type != ContentProperties_ImageProperties) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat(
+            "Expected ImageProperties for tensor ",
+            tensor_metadata.name() ? tensor_metadata.name()->str() : "#0",
+            ", got ", EnumNameContentProperties(type), "."),
+        TfLiteSupportStatus::kMetadataInvalidContentPropertiesError);
+  }
+
+  return tensor_metadata.content()->content_properties_as_ImageProperties();
+}
+
+StatusOr<absl::optional<NormalizationOptions>> GetNormalizationOptionsIfAny(
+    const TensorMetadata& tensor_metadata) {
+  ASSIGN_OR_RETURN(
+      const tflite::ProcessUnit* normalization_process_unit,
+      ModelMetadataExtractor::FindFirstProcessUnit(
+          tensor_metadata, tflite::ProcessUnitOptions_NormalizationOptions));
+  if (normalization_process_unit == nullptr) {
+    return {absl::nullopt};
+  }
+  const tflite::NormalizationOptions* tf_normalization_options =
+      normalization_process_unit->options_as_NormalizationOptions();
+  const auto mean_values = tf_normalization_options->mean();
+  const auto std_values = tf_normalization_options->std();
+  if (mean_values->size() != std_values->size()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat("NormalizationOptions: expected mean and std of same "
+                     "dimension, got ",
+                     mean_values->size(), " and ", std_values->size(), "."),
+        TfLiteSupportStatus::kMetadataInvalidProcessUnitsError);
+  }
+  absl::optional<NormalizationOptions> normalization_options;
+  if (mean_values->size() == 1) {
+    normalization_options = NormalizationOptions{
+        .mean_values = {mean_values->Get(0), mean_values->Get(0),
+                        mean_values->Get(0)},
+        .std_values = {std_values->Get(0), std_values->Get(0),
+                       std_values->Get(0)},
+        .num_values = 1};
+  } else if (mean_values->size() == 3) {
+    normalization_options = NormalizationOptions{
+        .mean_values = {mean_values->Get(0), mean_values->Get(1),
+                        mean_values->Get(2)},
+        .std_values = {std_values->Get(0), std_values->Get(1),
+                       std_values->Get(2)},
+        .num_values = 3};
+  } else {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat("NormalizationOptions: only 1 or 3 mean and std "
+                     "values are supported, got ",
+                     mean_values->size(), "."),
+        TfLiteSupportStatus::kMetadataInvalidProcessUnitsError);
+  }
+  return normalization_options;
+}
+
+}  // namespace
+
+StatusOr<ImageTensorSpecs> BuildInputImageTensorSpecs(
+    const TfLiteEngine::Interpreter& interpreter,
+    const tflite::metadata::ModelMetadataExtractor& metadata_extractor) {
+  ASSIGN_OR_RETURN(const TensorMetadata* metadata,
+                   GetInputTensorMetadataIfAny(metadata_extractor));
+
+  const ImageProperties* props = nullptr;
+  absl::optional<NormalizationOptions> normalization_options;
+  if (metadata != nullptr) {
+    ASSIGN_OR_RETURN(props, GetImagePropertiesIfAny(*metadata));
+    ASSIGN_OR_RETURN(normalization_options,
+                     GetNormalizationOptionsIfAny(*metadata));
+  }
+
+  if (TfLiteEngine::InputCount(&interpreter) != 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Models are assumed to have a single input.",
+        TfLiteSupportStatus::kInvalidNumInputTensorsError);
+  }
+
+  // Input-related specifications.
+  const TfLiteTensor* input_tensor = TfLiteEngine::GetInput(&interpreter, 0);
+  if (input_tensor->dims->size != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Only 4D tensors in BHWD layout are supported.",
+        TfLiteSupportStatus::kInvalidInputTensorDimensionsError);
+  }
+  static constexpr TfLiteType valid_types[] = {kTfLiteUInt8, kTfLiteFloat32};
+  TfLiteType input_type = input_tensor->type;
+  if (!absl::c_linear_search(valid_types, input_type)) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat(
+            "Type mismatch for input tensor ", input_tensor->name,
+            ". Requested one of these types: kTfLiteUint8/kTfLiteFloat32, got ",
+            TfLiteTypeGetName(input_type), "."),
+        TfLiteSupportStatus::kInvalidInputTensorTypeError);
+  }
+
+  // The expected layout is BHWD, i.e. batch x height x width x color
+  // See https://www.tensorflow.org/guide/tensors
+  const int batch = input_tensor->dims->data[0];
+  const int height = input_tensor->dims->data[1];
+  const int width = input_tensor->dims->data[2];
+  const int depth = input_tensor->dims->data[3];
+
+  if (props != nullptr && props->color_space() != ColorSpaceType_RGB) {
+    return CreateStatusWithPayload(StatusCode::kInvalidArgument,
+                                   "Only RGB color space is supported for now.",
+                                   TfLiteSupportStatus::kInvalidArgumentError);
+  }
+  if (batch != 1 || depth != 3) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrCat("The input tensor should have dimensions 1 x height x "
+                     "width x 3. Got ",
+                     batch, " x ", height, " x ", width, " x ", depth, "."),
+        TfLiteSupportStatus::kInvalidInputTensorDimensionsError);
+  }
+  int bytes_size = input_tensor->bytes;
+  size_t byte_depth =
+      input_type == kTfLiteFloat32 ? sizeof(float) : sizeof(uint8);
+
+  // Sanity checks.
+  if (input_type == kTfLiteFloat32) {
+    if (!normalization_options.has_value()) {
+      return CreateStatusWithPayload(
+          absl::StatusCode::kNotFound,
+          "Input tensor has type kTfLiteFloat32: it requires specifying "
+          "NormalizationOptions metadata to preprocess input images.",
+          TfLiteSupportStatus::kMetadataMissingNormalizationOptionsError);
+    } else if (bytes_size / sizeof(float) %
+                   normalization_options.value().num_values !=
+               0) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          "The number of elements in the input tensor must be a multiple of "
+          "the number of normalization parameters.",
+          TfLiteSupportStatus::kInvalidArgumentError);
+    }
+  }
+  if (width <= 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument, "The input width should be positive.",
+        TfLiteSupportStatus::kInvalidInputTensorDimensionsError);
+  }
+  if (height <= 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument, "The input height should be positive.",
+        TfLiteSupportStatus::kInvalidInputTensorDimensionsError);
+  }
+  if (bytes_size != height * width * depth * byte_depth) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "The input size in bytes does not correspond to the expected number of "
+        "pixels.",
+        TfLiteSupportStatus::kInvalidInputTensorSizeError);
+  }
+
+  // Note: in the future, additional checks against `props->default_size()`
+  // might be added. Also, verify that NormalizationOptions, if any, do specify
+  // a single value when color space is grayscale.
+
+  ImageTensorSpecs result;
+  result.image_width = width;
+  result.image_height = height;
+  result.color_space = ColorSpaceType_RGB;
+  result.tensor_type = input_type;
+  result.normalization_options = normalization_options;
+
+  return result;
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.h b/tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.h
new file mode 100644
index 0000000..536eed4
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/image_tensor_specs.h
@@ -0,0 +1,93 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_IMAGE_TENSOR_SPECS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_IMAGE_TENSOR_SPECS_H_
+
+#include <array>
+
+#include "absl/types/optional.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/tflite_engine.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Parameters used for input image normalization when input tensor has
+// kTfLiteFloat32 type.
+//
+// Exactly 1 or 3 values are expected for `mean_values` and `std_values`. In
+// case 1 value only is specified, it is used for all channels. E.g. for a RGB
+// image, the normalization is done as follow:
+//
+//   (R - mean_values[0]) / std_values[0]
+//   (G - mean_values[1]) / std_values[1]
+//   (B - mean_values[2]) / std_values[2]
+//
+// `num_values` keeps track of how many values have been provided, which should
+// be 1 or 3 (see above). In particular, single-channel grayscale images expect
+// only 1 value.
+struct NormalizationOptions {
+  std::array<float, 3> mean_values;
+  std::array<float, 3> std_values;
+  int num_values;
+};
+
+// Parameters related to the expected tensor specifications when the tensor
+// represents an image.
+//
+// E.g. input tensor specifications expected by the model at Invoke() time. In
+// such a case, and before running inference with the TF Lite interpreter, the
+// caller must use these values and perform image preprocessing and/or
+// normalization so as to fill the actual input tensor appropriately.
+struct ImageTensorSpecs {
+  // Expected image dimensions, e.g. image_width=224, image_height=224.
+  int image_width;
+  int image_height;
+  // Expected color space, e.g. color_space=RGB.
+  tflite::ColorSpaceType color_space;
+  // Expected input tensor type, e.g. if tensor_type=kTfLiteFloat32 the caller
+  // should usually perform some normalization to convert the uint8 pixels into
+  // floats (see NormalizationOptions in TF Lite Metadata for more details).
+  TfLiteType tensor_type;
+  // Optional normalization parameters read from TF Lite Metadata. Those are
+  // mandatory when tensor_type=kTfLiteFloat32 in order to convert the input
+  // image data into the expected range of floating point values, an error is
+  // returned otherwise (see sanity checks below). They should be ignored for
+  // other tensor input types, e.g. kTfLiteUInt8.
+  absl::optional<NormalizationOptions> normalization_options;
+};
+
+// Performs sanity checks on the expected input tensor including consistency
+// checks against model metadata, if any. For now, a single RGB input with BHWD
+// layout, where B = 1 and D = 3, is expected. Returns the corresponding input
+// specifications if they pass, or an error otherwise (too many input tensors,
+// etc).
+// Note: both interpreter and metadata extractor *must* be successfully
+// initialized before calling this function by means of (respectively):
+// - `tflite::InterpreterBuilder`,
+// - `tflite::metadata::ModelMetadataExtractor::CreateFromModelBuffer`.
+tflite::support::StatusOr<ImageTensorSpecs> BuildInputImageTensorSpecs(
+    const tflite::task::core::TfLiteEngine::Interpreter& interpreter,
+    const tflite::metadata::ModelMetadataExtractor& metadata_extractor);
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_IMAGE_TENSOR_SPECS_H_
diff --git a/tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.cc b/tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.cc
new file mode 100644
index 0000000..beb58eb
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.cc
@@ -0,0 +1,1499 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.h"
+
+#include <stdint.h>
+
+#include <memory>
+#include <string>
+
+#include "absl/status/status.h"
+#include "absl/strings/str_cat.h"
+#include "absl/strings/str_format.h"
+#include "include/libyuv.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+using ::absl::StatusCode;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::TfLiteSupportStatus;
+
+namespace {
+
+// Converts NV12 `buffer` to the `output_buffer` of the target color space.
+// Supported output format includes RGB24 and YV21.
+absl::Status ConvertFromNv12(const FrameBuffer& buffer,
+                             FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData yuv_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  switch (output_buffer->format()) {
+    case FrameBuffer::Format::kRGB: {
+      // The RAW format of Libyuv represents the 8-bit interleaved RGB format in
+      // the big endian style with R being the first byte in memory.
+      int ret = libyuv::NV12ToRAW(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.u_buffer,
+          yuv_data.uv_row_stride,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv NV12ToRAW operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kRGBA: {
+      // The libyuv ABGR format is interleaved RGBA format in memory.
+      int ret = libyuv::NV12ToABGR(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.u_buffer,
+          yuv_data.uv_row_stride,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv NV12ToABGR operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      int ret = libyuv::NV12ToI420(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.u_buffer,
+          yuv_data.uv_row_stride, const_cast<uint8_t*>(output_data.y_buffer),
+          output_data.y_row_stride, const_cast<uint8_t*>(output_data.u_buffer),
+          output_data.uv_row_stride, const_cast<uint8_t*>(output_data.v_buffer),
+          output_data.uv_row_stride, output_buffer->dimension().width,
+          output_buffer->dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv NV12ToI420 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kNV21: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      libyuv::CopyPlane(yuv_data.y_buffer, yuv_data.y_row_stride,
+                        const_cast<uint8*>(output_data.y_buffer),
+                        output_data.y_row_stride, buffer.dimension().width,
+                        buffer.dimension().height);
+      ASSIGN_OR_RETURN(
+          const FrameBuffer::Dimension uv_plane_dimension,
+          GetUvPlaneDimension(buffer.dimension(), buffer.format()));
+      libyuv::SwapUVPlane(yuv_data.u_buffer, yuv_data.uv_row_stride,
+                          const_cast<uint8*>(output_data.v_buffer),
+                          output_data.uv_row_stride, uv_plane_dimension.width,
+                          uv_plane_dimension.height);
+      break;
+    }
+    case FrameBuffer::Format::kGRAY: {
+      libyuv::CopyPlane(yuv_data.y_buffer, yuv_data.y_row_stride,
+                        const_cast<uint8*>(output_buffer->plane(0).buffer),
+                        output_buffer->plane(0).stride.row_stride_bytes,
+                        output_buffer->dimension().width,
+                        output_buffer->dimension().height);
+      break;
+    }
+    default:
+      return absl::InternalError(absl::StrFormat("Format %i is not supported.",
+                                                 output_buffer->format()));
+  }
+  return absl::OkStatus();
+}
+
+// Converts NV21 `buffer` into the `output_buffer` of the target color space.
+// Supported output format includes RGB24 and YV21.
+absl::Status ConvertFromNv21(const FrameBuffer& buffer,
+                             FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData yuv_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  switch (output_buffer->format()) {
+    case FrameBuffer::Format::kRGB: {
+      // The RAW format of Libyuv represents the 8-bit interleaved RGB format in
+      // the big endian style with R being the first byte in memory.
+      int ret = libyuv::NV21ToRAW(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.v_buffer,
+          yuv_data.uv_pixel_stride,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv NV21ToRAW operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kRGBA: {
+      // The libyuv ABGR format is interleaved RGBA format in memory.
+      int ret = libyuv::NV21ToABGR(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.v_buffer,
+          yuv_data.uv_pixel_stride,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv NV21ToABGR operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      int ret = libyuv::NV21ToI420(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.v_buffer,
+          yuv_data.uv_row_stride, const_cast<uint8_t*>(output_data.y_buffer),
+          output_data.y_row_stride, const_cast<uint8_t*>(output_data.u_buffer),
+          output_data.uv_row_stride, const_cast<uint8_t*>(output_data.v_buffer),
+          output_data.uv_row_stride, output_buffer->dimension().width,
+          output_buffer->dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv NV21ToI420 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kNV12: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      libyuv::CopyPlane(yuv_data.y_buffer, yuv_data.y_row_stride,
+                        const_cast<uint8*>(output_data.y_buffer),
+                        output_data.y_row_stride, buffer.dimension().width,
+                        buffer.dimension().height);
+      ASSIGN_OR_RETURN(
+          const FrameBuffer::Dimension uv_plane_dimension,
+          GetUvPlaneDimension(buffer.dimension(), buffer.format()));
+      libyuv::SwapUVPlane(yuv_data.v_buffer, yuv_data.uv_row_stride,
+                          const_cast<uint8*>(output_data.u_buffer),
+                          output_data.uv_row_stride, uv_plane_dimension.width,
+                          uv_plane_dimension.height);
+      break;
+    }
+    case FrameBuffer::Format::kGRAY: {
+      libyuv::CopyPlane(yuv_data.y_buffer, yuv_data.y_row_stride,
+                        const_cast<uint8*>(output_buffer->plane(0).buffer),
+                        output_buffer->plane(0).stride.row_stride_bytes,
+                        output_buffer->dimension().width,
+                        output_buffer->dimension().height);
+      break;
+    }
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.",
+                          output_buffer->format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+  return absl::OkStatus();
+}
+
+// Converts YV12/YV21 `buffer` to the `output_buffer` of the target color space.
+// Supported output format includes RGB24, NV12, and NV21.
+absl::Status ConvertFromYv(const FrameBuffer& buffer,
+                           FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData yuv_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  switch (output_buffer->format()) {
+    case FrameBuffer::Format::kRGB: {
+      // The RAW format of Libyuv represents the 8-bit interleaved RGB format in
+      // the big endian style with R being the first byte in memory.
+      int ret = libyuv::I420ToRAW(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.u_buffer,
+          yuv_data.uv_row_stride, yuv_data.v_buffer, yuv_data.uv_row_stride,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv I420ToRAW operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kRGBA: {
+      // The libyuv ABGR format is interleaved RGBA format in memory.
+      int ret = libyuv::I420ToABGR(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.u_buffer,
+          yuv_data.uv_row_stride, yuv_data.v_buffer, yuv_data.uv_row_stride,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv I420ToABGR operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kNV12: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      int ret = libyuv::I420ToNV12(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.u_buffer,
+          yuv_data.uv_row_stride, yuv_data.v_buffer, yuv_data.uv_row_stride,
+          const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+          const_cast<uint8*>(output_data.u_buffer), output_data.uv_row_stride,
+          output_buffer->dimension().width, output_buffer->dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv I420ToNV12 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kNV21: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      int ret = libyuv::I420ToNV21(
+          yuv_data.y_buffer, yuv_data.y_row_stride, yuv_data.u_buffer,
+          yuv_data.uv_row_stride, yuv_data.v_buffer, yuv_data.uv_row_stride,
+          const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+          const_cast<uint8*>(output_data.v_buffer), output_data.uv_row_stride,
+          output_buffer->dimension().width, output_buffer->dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv I420ToNV21 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kGRAY: {
+      libyuv::CopyPlane(yuv_data.y_buffer, yuv_data.y_row_stride,
+                        const_cast<uint8*>(output_buffer->plane(0).buffer),
+                        output_buffer->plane(0).stride.row_stride_bytes,
+                        output_buffer->dimension().width,
+                        output_buffer->dimension().height);
+      break;
+    }
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_yuv_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      ASSIGN_OR_RETURN(
+          const FrameBuffer::Dimension uv_plane_dimension,
+          GetUvPlaneDimension(buffer.dimension(), buffer.format()));
+      libyuv::CopyPlane(yuv_data.y_buffer, yuv_data.y_row_stride,
+                        const_cast<uint8*>(output_yuv_data.y_buffer),
+                        output_yuv_data.y_row_stride, buffer.dimension().width,
+                        buffer.dimension().height);
+      libyuv::CopyPlane(yuv_data.u_buffer, yuv_data.uv_row_stride,
+                        const_cast<uint8*>(output_yuv_data.u_buffer),
+                        output_yuv_data.uv_row_stride, uv_plane_dimension.width,
+                        uv_plane_dimension.height);
+      libyuv::CopyPlane(yuv_data.v_buffer, yuv_data.uv_row_stride,
+                        const_cast<uint8*>(output_yuv_data.v_buffer),
+                        output_yuv_data.uv_row_stride, uv_plane_dimension.width,
+                        uv_plane_dimension.height);
+      break;
+    }
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.",
+                          output_buffer->format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+  return absl::OkStatus();
+}
+
+// Resizes YV12/YV21 `buffer` to the target `output_buffer`.
+absl::Status ResizeYv(const FrameBuffer& buffer, FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  // TODO(b/151217096): Choose the optimal image resizing filter to optimize
+  // the model inference performance.
+  int ret = libyuv::I420Scale(
+      input_data.y_buffer, input_data.y_row_stride, input_data.u_buffer,
+      input_data.uv_row_stride, input_data.v_buffer, input_data.uv_row_stride,
+      buffer.dimension().width, buffer.dimension().height,
+      const_cast<uint8_t*>(output_data.y_buffer), output_data.y_row_stride,
+      const_cast<uint8_t*>(output_data.u_buffer), output_data.uv_row_stride,
+      const_cast<uint8_t*>(output_data.v_buffer), output_data.uv_row_stride,
+      output_buffer->dimension().width, output_buffer->dimension().height,
+      libyuv::FilterMode::kFilterBilinear);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv I420Scale operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Resizes NV12/NV21 `buffer` to the target `output_buffer`.
+absl::Status ResizeNv(const FrameBuffer& buffer, FrameBuffer* output_buffer) {
+  const int buffer_size =
+      GetFrameBufferByteSize(buffer.dimension(), FrameBuffer::Format::kYV21);
+  auto yuv_raw_buffer = absl::make_unique<uint8[]>(buffer_size);
+  ASSIGN_OR_RETURN(
+      std::unique_ptr<FrameBuffer> yuv_buffer,
+      CreateFromRawBuffer(yuv_raw_buffer.get(), buffer.dimension(),
+                          FrameBuffer::Format::kYV21, buffer.orientation()));
+  // TODO(b/151375918): Current implementation is a workaround by converting
+  // input NV12/NV21 buffer to the YV12 formats, resizing the YV12 buffer, and
+  // converting the resized YV12 buffer back to the target format. Consider
+  // optimizes this by adding the support of NV12/NV21 resizing in Libyuv.
+  if (buffer.format() == FrameBuffer::Format::kNV12) {
+    RETURN_IF_ERROR(ConvertFromNv12(buffer, yuv_buffer.get()));
+  } else if (buffer.format() == FrameBuffer::Format::kNV21) {
+    RETURN_IF_ERROR(ConvertFromNv21(buffer, yuv_buffer.get()));
+  } else {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Format %i is not supported.", buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  const int resized_buffer_size = GetFrameBufferByteSize(
+      output_buffer->dimension(), FrameBuffer::Format::kYV12);
+  auto resized_yuv_raw_buffer = absl::make_unique<uint8[]>(resized_buffer_size);
+  ASSIGN_OR_RETURN(std::unique_ptr<FrameBuffer> resized_yuv_buffer,
+                   CreateFromRawBuffer(resized_yuv_raw_buffer.get(),
+                                       output_buffer->dimension(),
+                                       FrameBuffer::Format::kYV12,
+                                       output_buffer->orientation()));
+  RETURN_IF_ERROR(ResizeYv(*yuv_buffer, resized_yuv_buffer.get()));
+
+  RETURN_IF_ERROR(ConvertFromYv(*resized_yuv_buffer, output_buffer));
+  return absl::OkStatus();
+}
+
+// Converts `buffer` to libyuv ARGB format and stores the conversion result
+// in `dest_argb`.
+absl::Status ConvertRgbToArgb(const FrameBuffer& buffer, uint8* dest_argb,
+                              int dest_stride_argb) {
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(buffer));
+  if (buffer.format() != FrameBuffer::Format::kRGB) {
+    return CreateStatusWithPayload(StatusCode::kInternal,
+                                   "RGB input format is expected.",
+                                   TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  if (dest_argb == nullptr || dest_stride_argb <= 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        "Invalid destination arguments for ConvertRgbToArgb.",
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+  int ret = libyuv::RGB24ToARGB(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      dest_argb, dest_stride_argb, buffer.dimension().width,
+      buffer.dimension().height);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv RGB24ToARGB operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Converts `src_argb` in libyuv ARGB format to FrameBuffer::kRGB format and
+// stores the conversion result in `output_buffer`.
+absl::Status ConvertArgbToRgb(uint8* src_argb, int src_stride_argb,
+                              FrameBuffer* output_buffer) {
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(*output_buffer));
+  if (output_buffer->format() != FrameBuffer::Format::kRGB) {
+    return absl::InternalError("RGB input format is expected.");
+  }
+
+  if (src_argb == nullptr || src_stride_argb <= 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal, "Invalid source arguments for ConvertArgbToRgb.",
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  if (output_buffer->plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        output_buffer->format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+  int ret = libyuv::ARGBToRGB24(
+      src_argb, src_stride_argb,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes,
+      output_buffer->dimension().width, output_buffer->dimension().height);
+
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv ARGBToRGB24 operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Converts `buffer` in FrameBuffer::kRGBA format to libyuv ARGB (BGRA in
+// memory) format and stores the conversion result in `dest_argb`.
+absl::Status ConvertRgbaToArgb(const FrameBuffer& buffer, uint8* dest_argb,
+                               int dest_stride_argb) {
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(buffer));
+  if (buffer.format() != FrameBuffer::Format::kRGBA) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal, "RGBA input format is expected.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  if (dest_argb == nullptr || dest_stride_argb <= 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        "Invalid source arguments for ConvertRgbaToArgb.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  int ret = libyuv::ABGRToARGB(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      dest_argb, dest_stride_argb, buffer.dimension().width,
+      buffer.dimension().height);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal, "Libyuv ABGRToARGB operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Converts kRGB `buffer` to the `output_buffer` of the target color space.
+absl::Status ConvertFromRgb(const FrameBuffer& buffer,
+                            FrameBuffer* output_buffer) {
+  if (output_buffer->format() == FrameBuffer::Format::kGRAY) {
+    int ret = libyuv::RAWToJ400(
+        buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+        const_cast<uint8*>(output_buffer->plane(0).buffer),
+        output_buffer->plane(0).stride.row_stride_bytes,
+        buffer.dimension().width, buffer.dimension().height);
+    if (ret != 0) {
+      return CreateStatusWithPayload(
+          StatusCode::kInternal, "Libyuv RAWToJ400 operation failed.",
+          TfLiteSupportStatus::kImageProcessingBackendError);
+    }
+    return absl::OkStatus();
+  } else if (output_buffer->format() == FrameBuffer::Format::kYV12 ||
+             output_buffer->format() == FrameBuffer::Format::kYV21 ||
+             output_buffer->format() == FrameBuffer::Format::kNV12 ||
+             output_buffer->format() == FrameBuffer::Format::kNV21) {
+    // libyuv does not support conversion directly from kRGB to kNV12 / kNV21.
+    // For kNV12 / kNV21, the implementation converts the kRGB to I420,
+    // then converts I420 to kNV12 / kNV21.
+    // TODO(b/153000936): use libyuv::RawToNV12 / libyuv::RawToNV21 when they
+    // are ready.
+    FrameBuffer::YuvData yuv_data;
+    std::unique_ptr<uint8[]> tmp_yuv_buffer;
+    std::unique_ptr<FrameBuffer> yuv_frame_buffer;
+    if (output_buffer->format() == FrameBuffer::Format::kNV12 ||
+        output_buffer->format() == FrameBuffer::Format::kNV21) {
+      tmp_yuv_buffer = absl::make_unique<uint8[]>(
+          GetFrameBufferByteSize(buffer.dimension(), output_buffer->format()));
+      ASSIGN_OR_RETURN(
+          yuv_frame_buffer,
+          CreateFromRawBuffer(tmp_yuv_buffer.get(), buffer.dimension(),
+                              FrameBuffer::Format::kYV21,
+                              output_buffer->orientation()));
+      ASSIGN_OR_RETURN(
+          yuv_data, FrameBuffer::GetYuvDataFromFrameBuffer(*yuv_frame_buffer));
+    } else {
+      ASSIGN_OR_RETURN(yuv_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+    }
+    int ret = libyuv::RAWToI420(
+        buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+        const_cast<uint8*>(yuv_data.y_buffer), yuv_data.y_row_stride,
+        const_cast<uint8*>(yuv_data.u_buffer), yuv_data.uv_row_stride,
+        const_cast<uint8*>(yuv_data.v_buffer), yuv_data.uv_row_stride,
+        buffer.dimension().width, buffer.dimension().height);
+    if (ret != 0) {
+      return CreateStatusWithPayload(
+          StatusCode::kInternal, "Libyuv RAWToI420 operation failed.",
+          TfLiteSupportStatus::kImageProcessingBackendError);
+    }
+    if (output_buffer->format() == FrameBuffer::Format::kNV12 ||
+        output_buffer->format() == FrameBuffer::Format::kNV21) {
+      return ConvertFromYv(*yuv_frame_buffer, output_buffer);
+    }
+    return absl::OkStatus();
+  }
+  return CreateStatusWithPayload(
+      StatusCode::kInternal,
+      absl::StrFormat("Format %i is not supported.", output_buffer->format()),
+      TfLiteSupportStatus::kImageProcessingError);
+}
+
+// Converts kRGBA `buffer` to the `output_buffer` of the target color space.
+absl::Status ConvertFromRgba(const FrameBuffer& buffer,
+                             FrameBuffer* output_buffer) {
+  switch (output_buffer->format()) {
+    case FrameBuffer::Format::kGRAY: {
+      // libyuv does not support convert kRGBA (ABGR) foramat. In this method,
+      // the implementation converts kRGBA format to ARGB and use ARGB buffer
+      // for conversion.
+      // TODO(b/141181395): Use libyuv::ABGRToJ400 when it is ready.
+
+      // Convert kRGBA to ARGB
+      int argb_buffer_size = GetFrameBufferByteSize(buffer.dimension(),
+                                                    FrameBuffer::Format::kRGBA);
+      auto argb_buffer = absl::make_unique<uint8[]>(argb_buffer_size);
+      const int argb_row_bytes = buffer.dimension().width * kRgbaPixelBytes;
+      RETURN_IF_ERROR(
+          ConvertRgbaToArgb(buffer, argb_buffer.get(), argb_row_bytes));
+
+      // Convert ARGB to kGRAY
+      int ret = libyuv::ARGBToJ400(
+          argb_buffer.get(), argb_row_bytes,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv ARGBToJ400 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kNV12: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      int ret = libyuv::ABGRToNV12(
+          buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+          const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+          const_cast<uint8*>(output_data.u_buffer), output_data.uv_row_stride,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv ABGRToNV12 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kNV21: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      int ret = libyuv::ABGRToNV21(
+          buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+          const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+          const_cast<uint8*>(output_data.v_buffer), output_data.uv_row_stride,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv ABGRToNV21 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21: {
+      ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                       FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+      int ret = libyuv::ABGRToI420(
+          buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+          const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+          const_cast<uint8*>(output_data.u_buffer), output_data.uv_row_stride,
+          const_cast<uint8*>(output_data.v_buffer), output_data.uv_row_stride,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv ABGRToI420 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    case FrameBuffer::Format::kRGB: {
+      // ARGB is BGRA in memory and RGB24 is BGR in memory. The removal of the
+      // alpha channel will not impact the RGB ordering.
+      int ret = libyuv::ARGBToRGB24(
+          buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+          const_cast<uint8*>(output_buffer->plane(0).buffer),
+          output_buffer->plane(0).stride.row_stride_bytes,
+          buffer.dimension().width, buffer.dimension().height);
+      if (ret != 0) {
+        return CreateStatusWithPayload(
+            StatusCode::kUnknown, "Libyuv ABGRToRGB24 operation failed.",
+            TfLiteSupportStatus::kImageProcessingBackendError);
+      }
+      break;
+    }
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Convert Rgba to format %i is not supported.",
+                          output_buffer->format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+  return absl::OkStatus();
+}
+
+// Returns libyuv rotation based on counter-clockwise angle_deg.
+libyuv::RotationMode GetLibyuvRotationMode(int angle_deg) {
+  switch (angle_deg) {
+    case 90:
+      return libyuv::kRotate270;
+    case 270:
+      return libyuv::kRotate90;
+    case 180:
+      return libyuv::kRotate180;
+    default:
+      return libyuv::kRotate0;
+  }
+}
+
+absl::Status RotateRgba(const FrameBuffer& buffer, int angle_deg,
+                        FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  // libyuv::ARGBRotate assumes RGBA buffer is in the interleaved format.
+  int ret = libyuv::ARGBRotate(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes, buffer.dimension().width,
+      buffer.dimension().height, GetLibyuvRotationMode(angle_deg % 360));
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv ARGBRotate operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+absl::Status RotateRgb(const FrameBuffer& buffer, int angle_deg,
+                       FrameBuffer* output_buffer) {
+  // libyuv does not support rotate kRGB (RGB24) foramat. In this method, the
+  // implementation converts kRGB format to ARGB and use ARGB buffer for
+  // rotation. The result is then convert back to RGB.
+
+  // Convert RGB to ARGB
+  int argb_buffer_size =
+      GetFrameBufferByteSize(buffer.dimension(), FrameBuffer::Format::kRGBA);
+  auto argb_buffer = absl::make_unique<uint8[]>(argb_buffer_size);
+  const int argb_row_bytes = buffer.dimension().width * kRgbaPixelBytes;
+  RETURN_IF_ERROR(ConvertRgbToArgb(buffer, argb_buffer.get(), argb_row_bytes));
+
+  // Rotate ARGB
+  auto argb_rotated_buffer = absl::make_unique<uint8[]>(argb_buffer_size);
+  int rotated_row_bytes = output_buffer->dimension().width * kRgbaPixelBytes;
+  // TODO(b/151954340): Optimize the current implementation by utilizing
+  // ARGBMirror for 180 degree rotation.
+  int ret = libyuv::ARGBRotate(
+      argb_buffer.get(), argb_row_bytes, argb_rotated_buffer.get(),
+      rotated_row_bytes, buffer.dimension().width, buffer.dimension().height,
+      GetLibyuvRotationMode(angle_deg % 360));
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv ARGBRotate operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  // Convert ARGB to RGB
+  return ConvertArgbToRgb(argb_rotated_buffer.get(), rotated_row_bytes,
+                          output_buffer);
+}
+
+absl::Status RotateGray(const FrameBuffer& buffer, int angle_deg,
+                        FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+  int ret = libyuv::RotatePlane(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes, buffer.dimension().width,
+      buffer.dimension().height, GetLibyuvRotationMode(angle_deg % 360));
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv RotatePlane operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Rotates YV12/YV21 frame buffer.
+absl::Status RotateYv(const FrameBuffer& buffer, int angle_deg,
+                      FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  int ret = libyuv::I420Rotate(
+      input_data.y_buffer, input_data.y_row_stride, input_data.u_buffer,
+      input_data.uv_row_stride, input_data.v_buffer, input_data.uv_row_stride,
+      const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+      const_cast<uint8*>(output_data.u_buffer), output_data.uv_row_stride,
+      const_cast<uint8*>(output_data.v_buffer), output_data.uv_row_stride,
+      buffer.dimension().width, buffer.dimension().height,
+      GetLibyuvRotationMode(angle_deg));
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv I420Rotate operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Rotates NV12/NV21 frame buffer.
+// TODO(b/152097364): Refactor NV12/NV21 rotation after libyuv explicitly
+// support that.
+absl::Status RotateNv(const FrameBuffer& buffer, int angle_deg,
+                      FrameBuffer* output_buffer) {
+  if (buffer.format() != FrameBuffer::Format::kNV12 &&
+      buffer.format() != FrameBuffer::Format::kNV21) {
+    return CreateStatusWithPayload(StatusCode::kInternal,
+                                   "kNV12 or kNV21 input formats are expected.",
+                                   TfLiteSupportStatus::kImageProcessingError);
+  }
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  const int rotated_buffer_size = GetFrameBufferByteSize(
+      output_buffer->dimension(), FrameBuffer::Format::kYV21);
+  auto rotated_yuv_raw_buffer = absl::make_unique<uint8[]>(rotated_buffer_size);
+  ASSIGN_OR_RETURN(std::unique_ptr<FrameBuffer> rotated_yuv_buffer,
+                   CreateFromRawBuffer(
+                       rotated_yuv_raw_buffer.get(), output_buffer->dimension(),
+                       /*target_format=*/FrameBuffer::Format::kYV21,
+                       output_buffer->orientation()));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData rotated_yuv_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*rotated_yuv_buffer));
+  // Get the first chroma plane and use it as the u plane. This is a workaround
+  // for optimizing NV21 rotation. For NV12, the implementation is logical
+  // correct. For NV21, use v plane as u plane will make the UV planes swapped
+  // in the intermediate rotated I420 frame. The output buffer is finally built
+  // by merging the swapped UV planes which produces V first interleaved UV
+  // buffer.
+  const uint8* chroma_buffer = buffer.format() == FrameBuffer::Format::kNV12
+                                   ? input_data.u_buffer
+                                   : input_data.v_buffer;
+  // Rotate the Y plane and store into the Y plane in `output_buffer`. Rotate
+  // the interleaved UV plane and store into the interleaved UV plane in
+  // `rotated_yuv_buffer`.
+  int ret = libyuv::NV12ToI420Rotate(
+      input_data.y_buffer, input_data.y_row_stride, chroma_buffer,
+      input_data.uv_row_stride, const_cast<uint8*>(output_data.y_buffer),
+      output_data.y_row_stride, const_cast<uint8*>(rotated_yuv_data.u_buffer),
+      rotated_yuv_data.uv_row_stride,
+      const_cast<uint8*>(rotated_yuv_data.v_buffer),
+      rotated_yuv_data.uv_row_stride, buffer.dimension().width,
+      buffer.dimension().height, GetLibyuvRotationMode(angle_deg % 360));
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv Nv12ToI420Rotate operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  // Merge rotated UV planes into the output buffer. For NV21, the UV buffer of
+  // the intermediate I420 frame is swapped. MergeUVPlane builds the interleaved
+  // VU buffer for NV21 by putting the U plane in the I420 frame which is
+  // actually the V plane from the input buffer first.
+  const uint8* output_chroma_buffer =
+      buffer.format() == FrameBuffer::Format::kNV12 ? output_data.u_buffer
+                                                    : output_data.v_buffer;
+  // The width and height arguments of `libyuv::MergeUVPlane()` represent the
+  // width and height of the UV planes.
+  libyuv::MergeUVPlane(
+      rotated_yuv_data.u_buffer, rotated_yuv_data.uv_row_stride,
+      rotated_yuv_data.v_buffer, rotated_yuv_data.uv_row_stride,
+      const_cast<uint8*>(output_chroma_buffer), output_data.uv_row_stride,
+      (output_buffer->dimension().width + 1) / 2,
+      (output_buffer->dimension().height + 1) / 2);
+  return absl::OkStatus();
+}
+
+// This method only supports kGRAY, kRGB, and kRGBA format.
+absl::Status FlipPlaneVertically(const FrameBuffer& buffer,
+                                 FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  ASSIGN_OR_RETURN(int pixel_stride, GetPixelStrides(buffer.format()));
+
+  // Flip vertically is achieved by passing in negative height.
+  libyuv::CopyPlane(buffer.plane(0).buffer,
+                    buffer.plane(0).stride.row_stride_bytes,
+                    const_cast<uint8*>(output_buffer->plane(0).buffer),
+                    output_buffer->plane(0).stride.row_stride_bytes,
+                    output_buffer->dimension().width * pixel_stride,
+                    -output_buffer->dimension().height);
+
+  return absl::OkStatus();
+}
+
+// This method only supports kGRAY, kRGBA, and kRGB formats.
+absl::Status CropPlane(const FrameBuffer& buffer, int x0, int y0, int x1,
+                       int y1, FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  ASSIGN_OR_RETURN(int pixel_stride, GetPixelStrides(buffer.format()));
+  FrameBuffer::Dimension crop_dimension = GetCropDimension(x0, x1, y0, y1);
+
+  // Cropping is achieved by adjusting origin to (x0, y0).
+  int adjusted_offset =
+      buffer.plane(0).stride.row_stride_bytes * y0 + x0 * pixel_stride;
+
+  libyuv::CopyPlane(buffer.plane(0).buffer + adjusted_offset,
+                    buffer.plane(0).stride.row_stride_bytes,
+                    const_cast<uint8*>(output_buffer->plane(0).buffer),
+                    output_buffer->plane(0).stride.row_stride_bytes,
+                    crop_dimension.width * pixel_stride, crop_dimension.height);
+
+  return absl::OkStatus();
+}
+
+// Crops NV12/NV21 FrameBuffer to the subregion defined by the top left pixel
+// position (x0, y0) and the bottom right pixel position (x1, y1).
+absl::Status CropNv(const FrameBuffer& buffer, int x0, int y0, int x1, int y1,
+                    FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  // Crop Y plane by copying the buffer with the origin offset to (x0, y0).
+  int crop_offset_y = input_data.y_row_stride * y0 + x0;
+  int crop_width = x1 - x0 + 1;
+  int crop_height = y1 - y0 + 1;
+  libyuv::CopyPlane(input_data.y_buffer + crop_offset_y,
+                    input_data.y_row_stride,
+                    const_cast<uint8*>(output_data.y_buffer),
+                    output_data.y_row_stride, crop_width, crop_height);
+  // Crop chroma plane by copying the buffer with the origin offset to
+  // (x0 / 2, y0 / 2);
+  // TODO(b/152629712): Investigate the impact of color shifting caused by the
+  // bounding box with odd X or Y starting positions.
+  int crop_offset_chroma = input_data.uv_row_stride * (y0 / 2) +
+                           input_data.uv_pixel_stride * (x0 / 2);
+  ASSIGN_OR_RETURN(const uint8* input_chroma_buffer, GetUvRawBuffer(buffer));
+  ASSIGN_OR_RETURN(const uint8* output_chroma_buffer,
+                   GetUvRawBuffer(*output_buffer));
+  libyuv::CopyPlane(
+      input_chroma_buffer + crop_offset_chroma, input_data.uv_row_stride,
+      const_cast<uint8*>(output_chroma_buffer), output_data.uv_row_stride,
+      /*width=*/(crop_width + 1) / 2 * 2, /*height=*/(crop_height + 1) / 2);
+  return absl::OkStatus();
+}
+
+// Crops YV12/YV21 FrameBuffer to the subregion defined by the top left pixel
+// position (x0, y0) and the bottom right pixel position (x1, y1).
+absl::Status CropYv(const FrameBuffer& buffer, int x0, int y0, int x1, int y1,
+                    FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  // Crop Y plane by copying the buffer with the origin offset to (x0, y0).
+  int crop_offset_y = input_data.y_row_stride * y0 + x0;
+  FrameBuffer::Dimension crop_dimension = GetCropDimension(x0, x1, y0, y1);
+  libyuv::CopyPlane(
+      input_data.y_buffer + crop_offset_y, input_data.y_row_stride,
+      const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+      crop_dimension.width, crop_dimension.height);
+  // Crop U plane by copying the buffer with the origin offset to
+  // (x0 / 2, y0 / 2).
+  ASSIGN_OR_RETURN(const FrameBuffer::Dimension crop_uv_dimension,
+                   GetUvPlaneDimension(crop_dimension, buffer.format()));
+  // TODO(b/152629712): Investigate the impact of color shifting caused by the
+  // bounding box with odd X or Y starting positions.
+  int crop_offset_chroma = input_data.uv_row_stride * (y0 / 2) +
+                           input_data.uv_pixel_stride * (x0 / 2);
+  libyuv::CopyPlane(
+      input_data.u_buffer + crop_offset_chroma, input_data.uv_row_stride,
+      const_cast<uint8*>(output_data.u_buffer), output_data.uv_row_stride,
+      crop_uv_dimension.width, crop_uv_dimension.height);
+  // Crop V plane by copying the buffer with the origin offset to
+  // (x0 / 2, y0 / 2);
+  libyuv::CopyPlane(
+      input_data.v_buffer + crop_offset_chroma, input_data.uv_row_stride,
+      const_cast<uint8*>(output_data.v_buffer), output_data.uv_row_stride,
+      /*width=*/(crop_dimension.width + 1) / 2,
+      /*height=*/(crop_dimension.height + 1) / 2);
+  return absl::OkStatus();
+}
+
+absl::Status CropResizeYuv(const FrameBuffer& buffer, int x0, int y0, int x1,
+                           int y1, FrameBuffer* output_buffer) {
+  FrameBuffer::Dimension crop_dimension = GetCropDimension(x0, x1, y0, y1);
+  if (crop_dimension == output_buffer->dimension()) {
+    switch (buffer.format()) {
+      case FrameBuffer::Format::kNV12:
+      case FrameBuffer::Format::kNV21:
+        return CropNv(buffer, x0, y0, x1, y1, output_buffer);
+      case FrameBuffer::Format::kYV12:
+      case FrameBuffer::Format::kYV21:
+        return CropYv(buffer, x0, y0, x1, y1, output_buffer);
+      default:
+        return CreateStatusWithPayload(
+            StatusCode::kInternal,
+            absl::StrFormat("Format %i is not supported.", buffer.format()),
+            TfLiteSupportStatus::kImageProcessingError);
+    }
+  }
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  // Cropping YUV planes by offsetting the origins of each plane.
+  // TODO(b/152629712): Investigate the impact of color shifting caused by the
+  // bounding box with odd X or Y starting positions.
+  const int plane_y_offset = input_data.y_row_stride * y0 + x0;
+  const int plane_uv_offset = input_data.uv_row_stride * (y0 / 2) +
+                              input_data.uv_pixel_stride * (x0 / 2);
+  FrameBuffer::Plane cropped_plane_y = {
+      /*buffer=*/input_data.y_buffer + plane_y_offset,
+      /*stride=*/{input_data.y_row_stride, /*pixel_stride_bytes=*/1}};
+  FrameBuffer::Plane cropped_plane_u = {
+      /*buffer=*/input_data.u_buffer + plane_uv_offset,
+      /*stride=*/{input_data.uv_row_stride, input_data.uv_pixel_stride}};
+  FrameBuffer::Plane cropped_plane_v = {
+      /*buffer=*/input_data.v_buffer + plane_uv_offset,
+      /*stride=*/{input_data.uv_row_stride, input_data.uv_pixel_stride}};
+
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kNV12: {
+      std::unique_ptr<FrameBuffer> cropped_buffer = FrameBuffer::Create(
+          {cropped_plane_y, cropped_plane_u, cropped_plane_v}, crop_dimension,
+          buffer.format(), buffer.orientation());
+      return ResizeNv(*cropped_buffer, output_buffer);
+    }
+    case FrameBuffer::Format::kNV21: {
+      std::unique_ptr<FrameBuffer> cropped_buffer = FrameBuffer::Create(
+          {cropped_plane_y, cropped_plane_v, cropped_plane_u}, crop_dimension,
+          buffer.format(), buffer.orientation());
+      return ResizeNv(*cropped_buffer, output_buffer);
+    }
+    case FrameBuffer::Format::kYV12: {
+      std::unique_ptr<FrameBuffer> cropped_buffer = FrameBuffer::Create(
+          {cropped_plane_y, cropped_plane_v, cropped_plane_u}, crop_dimension,
+          buffer.format(), buffer.orientation());
+      return ResizeYv(*cropped_buffer, output_buffer);
+    }
+    case FrameBuffer::Format::kYV21: {
+      std::unique_ptr<FrameBuffer> cropped_buffer = FrameBuffer::Create(
+          {cropped_plane_y, cropped_plane_u, cropped_plane_v}, crop_dimension,
+          buffer.format(), buffer.orientation());
+      return ResizeYv(*cropped_buffer, output_buffer);
+    }
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+  return absl::OkStatus();
+}
+
+absl::Status FlipHorizontallyRgba(const FrameBuffer& buffer,
+                                  FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  int ret = libyuv::ARGBMirror(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes,
+      output_buffer->dimension().width, output_buffer->dimension().height);
+
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv ARGBMirror operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  return absl::OkStatus();
+}
+
+// Flips `buffer` horizontally and store the result in `output_buffer`. This
+// method assumes all buffers have pixel stride equals to 1.
+absl::Status FlipHorizontallyPlane(const FrameBuffer& buffer,
+                                   FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+  libyuv::MirrorPlane(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes,
+      output_buffer->dimension().width, output_buffer->dimension().height);
+
+  return absl::OkStatus();
+}
+
+absl::Status ResizeRgb(const FrameBuffer& buffer, FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+  // libyuv doesn't support scale kRGB (RGB24) foramat. In this method,
+  // the implementation converts kRGB format to ARGB and use ARGB buffer for
+  // scaling. The result is then convert back to RGB.
+
+  // Convert RGB to ARGB
+  int argb_buffer_size =
+      GetFrameBufferByteSize(buffer.dimension(), FrameBuffer::Format::kRGBA);
+  auto argb_buffer = absl::make_unique<uint8[]>(argb_buffer_size);
+  const int argb_row_bytes = buffer.dimension().width * kRgbaPixelBytes;
+  RETURN_IF_ERROR(ConvertRgbToArgb(buffer, argb_buffer.get(), argb_row_bytes));
+
+  // Resize ARGB
+  int resized_argb_buffer_size = GetFrameBufferByteSize(
+      output_buffer->dimension(), FrameBuffer::Format::kRGBA);
+  auto resized_argb_buffer =
+      absl::make_unique<uint8[]>(resized_argb_buffer_size);
+  int resized_argb_row_bytes =
+      output_buffer->dimension().width * kRgbaPixelBytes;
+  int ret = libyuv::ARGBScale(
+      argb_buffer.get(), argb_row_bytes, buffer.dimension().width,
+      buffer.dimension().height, resized_argb_buffer.get(),
+      resized_argb_row_bytes, output_buffer->dimension().width,
+      output_buffer->dimension().height, libyuv::FilterMode::kFilterBilinear);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv ARGBScale operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  // Convert ARGB to RGB
+  return ConvertArgbToRgb(resized_argb_buffer.get(), resized_argb_row_bytes,
+                          output_buffer);
+}
+
+// Horizontally flip `buffer` and store the result in `output_buffer`.
+absl::Status FlipHorizontallyRgb(const FrameBuffer& buffer,
+                                 FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+
+#if LIBYUV_VERSION >= 1747
+  int ret = libyuv::RGB24Mirror(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes, buffer.dimension().width,
+      buffer.dimension().height);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv RGB24Mirror operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  return absl::OkStatus();
+#else
+#error LibyuvFrameBufferUtils requires LIBYUV_VERSION 1747 or above
+#endif  // LIBYUV_VERSION >= 1747
+}
+
+absl::Status ResizeRgba(const FrameBuffer& buffer, FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+  int ret = libyuv::ARGBScale(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      buffer.dimension().width, buffer.dimension().height,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes,
+      output_buffer->dimension().width, output_buffer->dimension().height,
+      libyuv::FilterMode::kFilterBilinear);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv ARGBScale operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Flips NV12/NV21 FrameBuffer horizontally.
+absl::Status FlipHorizontallyNv(const FrameBuffer& buffer,
+                                FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  ASSIGN_OR_RETURN(const uint8* input_chroma_buffer, GetUvRawBuffer(buffer));
+  ASSIGN_OR_RETURN(const uint8* output_chroma_buffer,
+                   GetUvRawBuffer(*output_buffer));
+
+  int ret = libyuv::NV12Mirror(
+      input_data.y_buffer, input_data.y_row_stride, input_chroma_buffer,
+      input_data.uv_row_stride, const_cast<uint8*>(output_data.y_buffer),
+      output_data.y_row_stride, const_cast<uint8*>(output_chroma_buffer),
+      output_data.uv_row_stride, buffer.dimension().width,
+      buffer.dimension().height);
+
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv NV12Mirror operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  return absl::OkStatus();
+}
+
+// Flips YV12/YV21 FrameBuffer horizontally.
+absl::Status FlipHorizontallyYv(const FrameBuffer& buffer,
+                                FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  int ret = libyuv::I420Mirror(
+      input_data.y_buffer, input_data.y_row_stride, input_data.u_buffer,
+      input_data.uv_row_stride, input_data.v_buffer, input_data.uv_row_stride,
+      const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+      const_cast<uint8*>(output_data.u_buffer), output_data.uv_row_stride,
+      const_cast<uint8*>(output_data.v_buffer), output_data.uv_row_stride,
+      buffer.dimension().width, buffer.dimension().height);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv I420Mirror operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+
+  return absl::OkStatus();
+}
+
+// Flips NV12/NV21 FrameBuffer vertically.
+absl::Status FlipVerticallyNv(const FrameBuffer& buffer,
+                              FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  // Flip Y plane vertically by passing a negative height.
+  libyuv::CopyPlane(input_data.y_buffer, input_data.y_row_stride,
+                    const_cast<uint8*>(output_data.y_buffer),
+                    output_data.y_row_stride, buffer.dimension().width,
+                    -output_buffer->dimension().height);
+  // Flip UV plane vertically by passing a negative height.
+  ASSIGN_OR_RETURN(const uint8* input_chroma_buffer, GetUvRawBuffer(buffer));
+  ASSIGN_OR_RETURN(const uint8* output_chroma_buffer,
+                   GetUvRawBuffer(*output_buffer));
+  ASSIGN_OR_RETURN(const FrameBuffer::Dimension uv_plane_dimension,
+                   GetUvPlaneDimension(buffer.dimension(), buffer.format()));
+  libyuv::CopyPlane(
+      input_chroma_buffer, input_data.uv_row_stride,
+      const_cast<uint8*>(output_chroma_buffer), output_data.uv_row_stride,
+      /*width=*/uv_plane_dimension.width * 2, -uv_plane_dimension.height);
+  return absl::OkStatus();
+}
+
+// Flips NV12/NV21 FrameBuffer vertically.
+absl::Status FlipVerticallyYv(const FrameBuffer& buffer,
+                              FrameBuffer* output_buffer) {
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData input_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(buffer));
+  ASSIGN_OR_RETURN(FrameBuffer::YuvData output_data,
+                   FrameBuffer::GetYuvDataFromFrameBuffer(*output_buffer));
+  // Flip buffer vertically by passing a negative height.
+  int ret = libyuv::I420Copy(
+      input_data.y_buffer, input_data.y_row_stride, input_data.u_buffer,
+      input_data.uv_row_stride, input_data.v_buffer, input_data.uv_row_stride,
+      const_cast<uint8*>(output_data.y_buffer), output_data.y_row_stride,
+      const_cast<uint8*>(output_data.u_buffer), output_data.uv_row_stride,
+      const_cast<uint8*>(output_data.v_buffer), output_data.uv_row_stride,
+      buffer.dimension().width, -buffer.dimension().height);
+  if (ret != 0) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown, "Libyuv I420Copy operation failed.",
+        TfLiteSupportStatus::kImageProcessingBackendError);
+  }
+  return absl::OkStatus();
+}
+
+// Resize `buffer` to metadata defined in `output_buffer`. This
+// method assumes buffer has pixel stride equals to 1 (grayscale equivalent).
+absl::Status ResizeGray(const FrameBuffer& buffer, FrameBuffer* output_buffer) {
+  if (buffer.plane_count() > 1) {
+    return CreateStatusWithPayload(
+        StatusCode::kInternal,
+        absl::StrFormat("Only single plane is supported for format %i.",
+                        buffer.format()),
+        TfLiteSupportStatus::kImageProcessingError);
+  }
+  libyuv::ScalePlane(
+      buffer.plane(0).buffer, buffer.plane(0).stride.row_stride_bytes,
+      buffer.dimension().width, buffer.dimension().height,
+      const_cast<uint8*>(output_buffer->plane(0).buffer),
+      output_buffer->plane(0).stride.row_stride_bytes,
+      output_buffer->dimension().width, output_buffer->dimension().height,
+      libyuv::FilterMode::kFilterBilinear);
+  return absl::OkStatus();
+}
+
+// This method only supports kGRAY, kRGBA, and kRGB formats.
+absl::Status CropResize(const FrameBuffer& buffer, int x0, int y0, int x1,
+                        int y1, FrameBuffer* output_buffer) {
+  FrameBuffer::Dimension crop_dimension = GetCropDimension(x0, x1, y0, y1);
+  if (crop_dimension == output_buffer->dimension()) {
+    return CropPlane(buffer, x0, y0, x1, y1, output_buffer);
+  }
+
+  ASSIGN_OR_RETURN(int pixel_stride, GetPixelStrides(buffer.format()));
+  // Cropping is achieved by adjusting origin to (x0, y0).
+  int adjusted_offset =
+      buffer.plane(0).stride.row_stride_bytes * y0 + x0 * pixel_stride;
+  FrameBuffer::Plane plane = {
+      /*buffer=*/buffer.plane(0).buffer + adjusted_offset,
+      /*stride=*/{buffer.plane(0).stride.row_stride_bytes, pixel_stride}};
+  auto adjusted_buffer =
+      FrameBuffer::Create({plane}, crop_dimension, buffer.format(),
+                          buffer.orientation(), buffer.timestamp());
+
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kRGB:
+      return ResizeRgb(*adjusted_buffer, output_buffer);
+    case FrameBuffer::Format::kRGBA:
+      return ResizeRgba(*adjusted_buffer, output_buffer);
+    case FrameBuffer::Format::kGRAY:
+      return ResizeGray(*adjusted_buffer, output_buffer);
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+}
+}  // namespace
+
+absl::Status LibyuvFrameBufferUtils::Crop(const FrameBuffer& buffer, int x0,
+                                          int y0, int x1, int y1,
+                                          FrameBuffer* output_buffer) {
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(buffer));
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(*output_buffer));
+  RETURN_IF_ERROR(
+      ValidateCropBufferInputs(buffer, *output_buffer, x0, y0, x1, y1));
+  RETURN_IF_ERROR(ValidateBufferFormats(buffer, *output_buffer));
+
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kRGBA:
+    case FrameBuffer::Format::kRGB:
+    case FrameBuffer::Format::kGRAY:
+      return CropResize(buffer, x0, y0, x1, y1, output_buffer);
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return CropResizeYuv(buffer, x0, y0, x1, y1, output_buffer);
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+}
+
+absl::Status LibyuvFrameBufferUtils::Resize(const FrameBuffer& buffer,
+                                            FrameBuffer* output_buffer) {
+  RETURN_IF_ERROR(ValidateResizeBufferInputs(buffer, *output_buffer));
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return ResizeYv(buffer, output_buffer);
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+      return ResizeNv(buffer, output_buffer);
+    case FrameBuffer::Format::kRGB:
+      return ResizeRgb(buffer, output_buffer);
+    case FrameBuffer::Format::kRGBA:
+      return ResizeRgba(buffer, output_buffer);
+    case FrameBuffer::Format::kGRAY:
+      return ResizeGray(buffer, output_buffer);
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+}
+
+absl::Status LibyuvFrameBufferUtils::Rotate(const FrameBuffer& buffer,
+                                            int angle_deg,
+                                            FrameBuffer* output_buffer) {
+  RETURN_IF_ERROR(
+      ValidateRotateBufferInputs(buffer, *output_buffer, angle_deg));
+  RETURN_IF_ERROR(ValidateBufferFormats(buffer, *output_buffer));
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(buffer));
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(*output_buffer));
+
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kGRAY:
+      return RotateGray(buffer, angle_deg, output_buffer);
+    case FrameBuffer::Format::kRGBA:
+      return RotateRgba(buffer, angle_deg, output_buffer);
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+      return RotateNv(buffer, angle_deg, output_buffer);
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return RotateYv(buffer, angle_deg, output_buffer);
+    case FrameBuffer::Format::kRGB:
+      return RotateRgb(buffer, angle_deg, output_buffer);
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+}
+
+absl::Status LibyuvFrameBufferUtils::FlipHorizontally(
+    const FrameBuffer& buffer, FrameBuffer* output_buffer) {
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(buffer));
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(*output_buffer));
+  RETURN_IF_ERROR(ValidateFlipBufferInputs(buffer, *output_buffer));
+  RETURN_IF_ERROR(ValidateBufferFormats(buffer, *output_buffer));
+
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kRGBA:
+      return FlipHorizontallyRgba(buffer, output_buffer);
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return FlipHorizontallyYv(buffer, output_buffer);
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+      return FlipHorizontallyNv(buffer, output_buffer);
+    case FrameBuffer::Format::kRGB:
+      return FlipHorizontallyRgb(buffer, output_buffer);
+    case FrameBuffer::Format::kGRAY:
+      return FlipHorizontallyPlane(buffer, output_buffer);
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+}
+
+absl::Status LibyuvFrameBufferUtils::FlipVertically(
+    const FrameBuffer& buffer, FrameBuffer* output_buffer) {
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(buffer));
+  RETURN_IF_ERROR(ValidateBufferPlaneMetadata(*output_buffer));
+  RETURN_IF_ERROR(ValidateFlipBufferInputs(buffer, *output_buffer));
+  RETURN_IF_ERROR(ValidateBufferFormats(buffer, *output_buffer));
+
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kRGBA:
+    case FrameBuffer::Format::kRGB:
+    case FrameBuffer::Format::kGRAY:
+      return FlipPlaneVertically(buffer, output_buffer);
+    case FrameBuffer::Format::kNV12:
+    case FrameBuffer::Format::kNV21:
+      return FlipVerticallyNv(buffer, output_buffer);
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return FlipVerticallyYv(buffer, output_buffer);
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+}
+
+absl::Status LibyuvFrameBufferUtils::Convert(const FrameBuffer& buffer,
+                                             FrameBuffer* output_buffer) {
+  RETURN_IF_ERROR(
+      ValidateConvertFormats(buffer.format(), output_buffer->format()));
+  switch (buffer.format()) {
+    case FrameBuffer::Format::kNV12:
+      return ConvertFromNv12(buffer, output_buffer);
+    case FrameBuffer::Format::kNV21:
+      return ConvertFromNv21(buffer, output_buffer);
+    case FrameBuffer::Format::kYV12:
+    case FrameBuffer::Format::kYV21:
+      return ConvertFromYv(buffer, output_buffer);
+    case FrameBuffer::Format::kRGB:
+      return ConvertFromRgb(buffer, output_buffer);
+    case FrameBuffer::Format::kRGBA:
+      return ConvertFromRgba(buffer, output_buffer);
+    default:
+      return CreateStatusWithPayload(
+          StatusCode::kInternal,
+          absl::StrFormat("Format %i is not supported.", buffer.format()),
+          TfLiteSupportStatus::kImageProcessingError);
+  }
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.h b/tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.h
new file mode 100644
index 0000000..0d001c8
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/libyuv_frame_buffer_utils.h
@@ -0,0 +1,76 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_LIBYUV_FRAME_BUFFER_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_LIBYUV_FRAME_BUFFER_UTILS_H_
+
+#include "absl/status/status.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_utils_interface.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Libyuv image processing engine conforms to FrameBufferUtilsInterface.
+// Although this class provides public APIs, it is recommended to use the public
+// APIs defined in frame_buffer_utils.h for higher level abstraction and better
+// functionality support.
+class LibyuvFrameBufferUtils : public FrameBufferUtilsInterface {
+ public:
+  LibyuvFrameBufferUtils() = default;
+  ~LibyuvFrameBufferUtils() override = default;
+
+  // Crops input `buffer` to the specified subregions and resizes the cropped
+  // region to the target image resolution defined by the `output_buffer`.
+  //
+  // (x0, y0) represents the top-left point of the buffer.
+  // (x1, y1) represents the bottom-right point of the buffer.
+  //
+  // Crop region dimensions must be equal or smaller than input `buffer`
+  // dimensions.
+  absl::Status Crop(const FrameBuffer& buffer, int x0, int y0, int x1, int y1,
+                    FrameBuffer* output_buffer) override;
+
+  // Resizes `buffer` to the size of the given `output_buffer`.
+  absl::Status Resize(const FrameBuffer& buffer,
+                      FrameBuffer* output_buffer) override;
+
+  // Rotates `buffer` counter-clockwise by the given `angle_deg` (in degrees).
+  //
+  // The given angle must be a multiple of 90 degrees.
+  absl::Status Rotate(const FrameBuffer& buffer, int angle_deg,
+                      FrameBuffer* output_buffer) override;
+
+  // Flips `buffer` horizontally.
+  absl::Status FlipHorizontally(const FrameBuffer& buffer,
+                                FrameBuffer* output_buffer) override;
+
+  // Flips `buffer` vertically.
+  absl::Status FlipVertically(const FrameBuffer& buffer,
+                              FrameBuffer* output_buffer) override;
+
+  // Converts `buffer`'s format to the format of the given `output_buffer`.
+  //
+  // Grayscale format cannot be converted to other formats.
+  absl::Status Convert(const FrameBuffer& buffer,
+                       FrameBuffer* output_buffer) override;
+};
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_LIBYUV_FRAME_BUFFER_UTILS_H_
diff --git a/tensorflow_lite_support/cc/task/vision/utils/score_calibration.cc b/tensorflow_lite_support/cc/task/vision/utils/score_calibration.cc
new file mode 100644
index 0000000..773ab76
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/score_calibration.cc
@@ -0,0 +1,225 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/cc/task/vision/utils/score_calibration.h"
+
+#include <cmath>
+#include <memory>
+#include <utility>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/str_split.h"
+#include "absl/strings/string_view.h"
+#include "absl/types/optional.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+namespace {
+
+using ::absl::StatusCode;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+
+// Used to prevent log(<=0.0) in ClampedLog() calls.
+constexpr float kLogScoreMinimum = 1e-16;
+
+// Returns the following, depending on x:
+//   x => threshold: log(x)
+//   x < threshold: 2 * log(thresh) - log(2 * thresh - x)
+// This form (a) is anti-symmetric about the threshold and (b) has continuous
+// value and first derivative. This is done to prevent taking the log of values
+// close to 0 which can lead to floating point errors and is better than simple
+// clamping since it preserves order for scores less than the threshold.
+float ClampedLog(float x, float threshold) {
+  if (x < threshold) {
+    return 2.0 * std::log(static_cast<double>(threshold)) -
+           log(2.0 * threshold - x);
+  }
+  return std::log(static_cast<double>(x));
+}
+
+// Applies the specified score transformation to the provided score.
+// Currently supports the following,
+//   IDENTITY         : f(x) = x
+//   LOG              : f(x) = log(x)
+//   INVERSE_LOGISTIC : f(x) = log(x) - log(1-x)
+float ApplyScoreTransformation(float score, const ScoreTransformation& type) {
+  switch (type) {
+    case ScoreTransformation::kIDENTITY:
+      return score;
+    case ScoreTransformation::kINVERSE_LOGISTIC:
+      return (ClampedLog(score, kLogScoreMinimum) -
+              ClampedLog(1.0 - score, kLogScoreMinimum));
+    case ScoreTransformation::kLOG:
+      return ClampedLog(score, kLogScoreMinimum);
+  }
+}
+
+// Builds a single Sigmoid from the label name and associated CSV file line.
+StatusOr<Sigmoid> SigmoidFromLabelAndLine(absl::string_view label,
+                                          absl::string_view line) {
+  std::vector<absl::string_view> str_params = absl::StrSplit(line, ',');
+  if (str_params.size() != 3 && str_params.size() != 4) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Expected 3 or 4 parameters per line in score "
+                        "calibration file, got %d.",
+                        str_params.size()),
+        TfLiteSupportStatus::kMetadataMalformedScoreCalibrationError);
+  }
+  std::vector<float> float_params(4);
+  for (int i = 0; i < str_params.size(); ++i) {
+    if (!absl::SimpleAtof(str_params[i], &float_params[i])) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat(
+              "Could not parse score calibration parameter as float: %s.",
+              str_params[i]),
+          TfLiteSupportStatus::kMetadataMalformedScoreCalibrationError);
+    }
+  }
+  Sigmoid sigmoid;
+  sigmoid.label = std::string(label);
+  sigmoid.scale = float_params[0];
+  sigmoid.slope = float_params[1];
+  sigmoid.offset = float_params[2];
+  if (str_params.size() == 4) {
+    sigmoid.min_uncalibrated_score = float_params[3];
+  }
+  return sigmoid;
+}
+
+// Converts a tflite::ScoreTransformationType to its
+// tflite::task::vision::ScoreTransformation equivalent.
+ScoreTransformation ConvertScoreTransformationType(
+    tflite::ScoreTransformationType type) {
+  switch (type) {
+    case tflite::ScoreTransformationType_IDENTITY:
+      return ScoreTransformation::kIDENTITY;
+    case tflite::ScoreTransformationType_LOG:
+      return ScoreTransformation::kLOG;
+    case tflite::ScoreTransformationType_INVERSE_LOGISTIC:
+      return ScoreTransformation::kINVERSE_LOGISTIC;
+  }
+}
+
+}  // namespace
+
+std::ostream& operator<<(std::ostream& os, const Sigmoid& s) {
+  os << s.label << "," << s.slope << "," << s.offset << "," << s.scale;
+  if (s.min_uncalibrated_score.has_value()) {
+    os << "," << s.min_uncalibrated_score.value();
+  }
+  return os;
+}
+
+ScoreCalibration::ScoreCalibration() {}
+ScoreCalibration::~ScoreCalibration() {}
+
+absl::Status ScoreCalibration::InitializeFromParameters(
+    const SigmoidCalibrationParameters& params) {
+  sigmoid_parameters_ = std::move(params);
+  // Fill in the map from label -> sigmoid.
+  sigmoid_parameters_map_.clear();
+  for (const auto& sigmoid : sigmoid_parameters_.sigmoid) {
+    sigmoid_parameters_map_.insert_or_assign(sigmoid.label, sigmoid);
+  }
+  return absl::OkStatus();
+}
+
+float ScoreCalibration::ComputeCalibratedScore(const std::string& label,
+                                               float uncalibrated_score) const {
+  absl::optional<Sigmoid> sigmoid = FindSigmoidParameters(label);
+  if (!sigmoid.has_value() ||
+      (sigmoid.value().min_uncalibrated_score.has_value() &&
+       uncalibrated_score < sigmoid.value().min_uncalibrated_score.value())) {
+    return sigmoid_parameters_.default_score;
+  }
+
+  float transformed_score = ApplyScoreTransformation(
+      uncalibrated_score, sigmoid_parameters_.score_transformation);
+  float scale_shifted_score =
+      transformed_score * sigmoid.value().slope + sigmoid.value().offset;
+
+  // For numerical stability use 1 / (1+exp(-x)) when scale_shifted_score >= 0
+  // and exp(x) / (1+exp(x)) when scale_shifted_score < 0.
+  if (scale_shifted_score >= 0.0) {
+    return sigmoid.value().scale /
+           (1.0 + std::exp(static_cast<double>(-scale_shifted_score)));
+  } else {
+    float score_exp = std::exp(static_cast<double>(scale_shifted_score));
+    return sigmoid.value().scale * score_exp / (1.0 + score_exp);
+  }
+}
+
+absl::optional<Sigmoid> ScoreCalibration::FindSigmoidParameters(
+    const std::string& label) const {
+  auto it = sigmoid_parameters_map_.find(label);
+  if (it != sigmoid_parameters_map_.end()) {
+    return it->second;
+  } else if (sigmoid_parameters_.default_sigmoid.has_value()) {
+    return sigmoid_parameters_.default_sigmoid.value();
+  }
+  return absl::nullopt;
+}
+
+StatusOr<SigmoidCalibrationParameters> BuildSigmoidCalibrationParams(
+    const tflite::ScoreCalibrationOptions& score_calibration_options,
+    absl::string_view score_calibration_file,
+    const std::vector<LabelMapItem>& label_map_items) {
+  // Split file lines and perform sanity checks.
+  if (score_calibration_file.empty()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "Expected non-empty score calibration file.");
+  }
+  std::vector<absl::string_view> lines =
+      absl::StrSplit(score_calibration_file, '\n');
+  if (label_map_items.size() != lines.size()) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        absl::StrFormat("Mismatch between number of labels (%d) and score "
+                        "calibration parameters (%d).",
+                        label_map_items.size(), lines.size()),
+        TfLiteSupportStatus::kMetadataNumLabelsMismatchError);
+  }
+  // Initialize SigmoidCalibrationParameters with its class-agnostic parameters.
+  SigmoidCalibrationParameters sigmoid_params = {};
+  sigmoid_params.score_transformation = ConvertScoreTransformationType(
+      score_calibration_options.score_transformation());
+  sigmoid_params.default_score = score_calibration_options.default_score();
+  std::vector<Sigmoid> sigmoid_vector;
+  // Fill sigmoids for each class with parameters in the file.
+  for (int i = 0; i < label_map_items.size(); ++i) {
+    if (lines[i].empty()) {
+      continue;
+    }
+    ASSIGN_OR_RETURN(Sigmoid sigmoid, SigmoidFromLabelAndLine(
+                                          label_map_items[i].name, lines[i]));
+    sigmoid_vector.emplace_back(std::move(sigmoid));
+  }
+  sigmoid_params.sigmoid = std::move(sigmoid_vector);
+
+  return sigmoid_params;
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/task/vision/utils/score_calibration.h b/tensorflow_lite_support/cc/task/vision/utils/score_calibration.h
new file mode 100644
index 0000000..c3f0bf8
--- /dev/null
+++ b/tensorflow_lite_support/cc/task/vision/utils/score_calibration.h
@@ -0,0 +1,146 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_SCORE_CALIBRATION_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_SCORE_CALIBRATION_H_
+
+#include <iostream>
+#include <map>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/container/flat_hash_map.h"
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+#include "absl/types/optional.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/vision/core/label_map_item.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Sigmoid structure.
+struct Sigmoid {
+  Sigmoid() : scale(1.0) {}
+  Sigmoid(std::string label, float slope, float offset, float scale = 1.0,
+          absl::optional<float> min_uncalibrated_score = absl::nullopt)
+      : label(label),
+        slope(slope),
+        offset(offset),
+        scale(scale),
+        min_uncalibrated_score(min_uncalibrated_score) {}
+
+  bool operator==(const Sigmoid& other) const {
+    return label == other.label && slope == other.slope &&
+           offset == other.offset && scale == other.scale &&
+           min_uncalibrated_score == other.min_uncalibrated_score;
+  }
+
+  // Unique label corresponding to the sigmoid parameters.
+  std::string label;
+  float slope;
+  float offset;
+  float scale;
+  absl::optional<float> min_uncalibrated_score;
+};
+
+std::ostream& operator<<(std::ostream& os, const Sigmoid& s);
+
+// Transformation function to use for computing transformation scores.
+enum class ScoreTransformation {
+  kIDENTITY,         // f(x) = x
+  kLOG,              // f(x) = log(x)
+  kINVERSE_LOGISTIC  // f(x) = log(x) - log(1 - x)
+};
+
+// Sigmoid calibration parameters.
+struct SigmoidCalibrationParameters {
+  SigmoidCalibrationParameters()
+      : default_score(0.0),
+        score_transformation(ScoreTransformation::kIDENTITY) {}
+  explicit SigmoidCalibrationParameters(
+      std::vector<Sigmoid> sigmoid,
+      ScoreTransformation score_transformation = ScoreTransformation::kIDENTITY,
+      absl::optional<Sigmoid> default_sigmoid = absl::nullopt,
+      float default_score = 0.0)
+      : sigmoid(sigmoid),
+        default_sigmoid(default_sigmoid),
+        default_score(default_score),
+        score_transformation(score_transformation) {}
+  // A vector of Sigmoid associated to the ScoreCalibration instance.
+  std::vector<Sigmoid> sigmoid;
+  // If set, this sigmoid will be applied to any non-matching labels.
+  absl::optional<Sigmoid> default_sigmoid;
+  // The default score for non-matching labels. Only used if default_sigmoid
+  // isn't set.
+  float default_score;
+  // Function for computing a transformation score prior to sigmoid fitting.
+  ScoreTransformation score_transformation;
+};
+
+// This class is used to calibrate predicted scores so that scores are
+// comparable across labels. Depending on the particular calibration parameters
+// being used, the calibrated scores can also be approximately interpreted as a
+// likelihood of being correct. For a given TF Lite model, such parameters are
+// typically obtained from TF Lite Metadata (see ScoreCalibrationOptions).
+class ScoreCalibration {
+ public:
+  ScoreCalibration();
+  ~ScoreCalibration();
+
+  // Transfers input parameters and construct a label to sigmoid map.
+  absl::Status InitializeFromParameters(
+      const SigmoidCalibrationParameters& params);
+
+  // Returns a calibrated score given a label string and uncalibrated score. The
+  // calibrated score will be in the range [0.0, 1.0] and can loosely be
+  // interpreted as a likelihood of the label being correct.
+  float ComputeCalibratedScore(const std::string& label,
+                               float uncalibrated_score) const;
+
+ private:
+  // Finds the sigmoid parameters corresponding to the provided label.
+  absl::optional<Sigmoid> FindSigmoidParameters(const std::string& label) const;
+
+  // Parameters for internal states.
+  SigmoidCalibrationParameters sigmoid_parameters_;
+
+  // Maps label strings to the particular sigmoid stored in sigmoid_parameters_.
+  absl::flat_hash_map<std::string, Sigmoid> sigmoid_parameters_map_;
+};
+
+// Builds SigmoidCalibrationParameters using data obtained from TF Lite Metadata
+// (see ScoreCalibrationOptions in metadata schema).
+//
+// The provided `score_calibration_file` represents the contents of the score
+// calibration associated file (TENSOR_AXIS_SCORE_CALIBRATION), i.e. one set of
+// parameters (scale, slope, etc) per line. Each line must be in 1:1
+// correspondence with `label_map_items`, so as to associate each sigmoid to its
+// corresponding label name. Returns an error if no valid parameters could be
+// built (e.g. malformed parameters).
+tflite::support::StatusOr<SigmoidCalibrationParameters>
+BuildSigmoidCalibrationParams(
+    const tflite::ScoreCalibrationOptions& score_calibration_options,
+    absl::string_view score_calibration_file,
+    const std::vector<LabelMapItem>& label_map_items);
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_UTILS_SCORE_CALIBRATION_H_
diff --git a/tensorflow_lite_support/cc/text/tokenizers/BUILD b/tensorflow_lite_support/cc/text/tokenizers/BUILD
new file mode 100644
index 0000000..3ad8da2
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/BUILD
@@ -0,0 +1,191 @@
+# This package contains C++ support libraries that Java libraries can invoke.
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load(
+    "@org_tensorflow//tensorflow/lite:build_def.bzl",
+    "tflite_copts",
+    "tflite_jni_binary",
+)
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "tokenizer",
+    hdrs = [
+        "tokenizer.h",
+    ],
+    deps = [
+        "@com_google_absl//absl/strings",
+    ],
+)
+
+cc_library(
+    name = "tokenizer_jni_lib",
+    srcs = [
+        "tokenizer_jni_lib.cc",
+    ],
+    hdrs = [
+        "tokenizer_jni_lib.h",
+    ],
+    deps = [
+        ":tokenizer",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "@org_tensorflow//tensorflow/lite/java/jni",
+    ],
+)
+
+cc_library(
+    name = "bert_tokenizer",
+    srcs = [
+        "bert_tokenizer.cc",
+    ],
+    hdrs = [
+        "bert_tokenizer.h",
+    ],
+    deps = [
+        ":tokenizer",
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/utils:common_utils",
+        "@com_google_absl//absl/container:flat_hash_map",
+        "@com_googlesource_code_re2//:re2",
+        "@org_tensorflow_text//tensorflow_text/core/kernels:regex_split",
+        "@org_tensorflow_text//tensorflow_text/core/kernels:wordpiece_tokenizer",
+    ],
+)
+
+cc_library(
+    name = "bert_tokenizer_jni_lib",
+    srcs = [
+        "bert_tokenizer_jni.cc",
+    ],
+    copts = tflite_copts(),
+    linkopts = [
+        "-lm",
+        "-ldl",
+    ],
+    deps = [
+        ":bert_tokenizer",
+        ":tokenizer_jni_lib",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "@com_google_absl//absl/memory",
+        "@org_tensorflow//tensorflow/lite/java/jni",
+    ],
+    alwayslink = 1,
+)
+
+tflite_jni_binary(
+    name = "libbert_tokenizer_jni.so",
+    deps = [
+        ":bert_tokenizer_jni_lib",
+    ],
+)
+
+cc_library(
+    name = "bert_tokenizer_runtime",
+    srcs = ["libbert_tokenizer_jni.so"],
+    alwayslink = 1,
+)
+
+android_library(
+    name = "bert_tokenizer_jni",
+    custom_package = "org.tensorflow.lite.support.text",
+    manifest = "DummyManifest.xml",
+    resource_files = [],
+    deps = [
+        ":bert_tokenizer_runtime",  # build_cleaner: skip
+    ],
+)
+
+cc_library(
+    name = "sentencepiece_tokenizer",
+    hdrs = [
+        "sentencepiece_tokenizer.h",
+    ],
+    deps = [
+        ":tokenizer",
+        "@com_google_sentencepiece//src:sentencepiece_processor",
+    ],
+)
+
+cc_library(
+    name = "sentencepiece_jni_lib",
+    srcs = [
+        "sentencepiece_jni.cc",
+    ],
+    copts = tflite_copts(),
+    linkopts = [
+        "-lm",
+        "-ldl",
+    ],
+    deps = [
+        ":sentencepiece_tokenizer",
+        ":tokenizer_jni_lib",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/strings",
+        "@org_tensorflow//tensorflow/lite/java/jni",
+    ],
+    alwayslink = 1,
+)
+
+cc_library(
+    name = "sentencepiece_runtime",
+    srcs = ["libsentencepiece_jni.so"],
+    alwayslink = 1,
+)
+
+tflite_jni_binary(
+    name = "libsentencepiece_jni.so",
+    deps = [
+        ":sentencepiece_jni_lib",
+    ],
+)
+
+android_library(
+    name = "sentencepiece_jni",
+    custom_package = "org.tensorflow.lite.support.text",
+    manifest = "DummyManifest.xml",
+    resource_files = [],
+    deps = [
+        ":sentencepiece_runtime",  # build_cleaner: skip
+    ],
+)
+
+cc_library(
+    name = "tokenizer_utils",
+    srcs = ["tokenizer_utils.cc"],
+    hdrs = [
+        "tokenizer_utils.h",
+    ],
+    deps = [
+        ":bert_tokenizer",
+        ":regex_tokenizer",
+        ":sentencepiece_tokenizer",
+        ":tokenizer",
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "//tensorflow_lite_support/metadata/cc:metadata_extractor",
+        "@com_google_absl//absl/status",
+    ],
+)
+
+cc_library(
+    name = "regex_tokenizer",
+    srcs = [
+        "regex_tokenizer.cc",
+    ],
+    hdrs = [
+        "regex_tokenizer.h",
+    ],
+    deps = [
+        ":tokenizer",
+        "//tensorflow_lite_support/cc/utils:common_utils",
+        "@com_google_absl//absl/container:node_hash_map",
+        "@com_google_absl//absl/strings",
+        "@com_googlesource_code_re2//:re2",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/text/tokenizers/DummyManifest.xml b/tensorflow_lite_support/cc/text/tokenizers/DummyManifest.xml
new file mode 100644
index 0000000..ff02507
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/DummyManifest.xml
@@ -0,0 +1,19 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!--
+ Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+     http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+-->
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="org.tensorflow.lite.support.text">
+</manifest>
diff --git a/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.cc b/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.cc
new file mode 100644
index 0000000..aeb887c
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.cc
@@ -0,0 +1,108 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h"
+
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+FlatHashMapBackedWordpiece::FlatHashMapBackedWordpiece(
+    const std::vector<std::string>& vocab)
+    : vocab_{vocab} {
+  for (int i = 0; i < vocab_.size(); ++i) {
+    index_map_[vocab_[i]] = i;
+  }
+}
+
+tensorflow::text::LookupStatus FlatHashMapBackedWordpiece::Contains(
+    absl::string_view key, bool* value) const {
+  *value = index_map_.contains(key);
+  return tensorflow::text::LookupStatus();
+}
+
+bool FlatHashMapBackedWordpiece::LookupId(const absl::string_view key,
+                                          int* result) const {
+  auto it = index_map_.find(key);
+  if (it == index_map_.end()) {
+    return false;
+  }
+  *result = it->second;
+  return true;
+}
+
+bool FlatHashMapBackedWordpiece::LookupWord(int vocab_id,
+                                            absl::string_view* result) const {
+  if (vocab_id >= vocab_.size() || vocab_id < 0) {
+    return false;
+  }
+  *result = vocab_[vocab_id];
+  return true;
+}
+
+TokenizerResult BertTokenizer::Tokenize(const std::string& input) {
+  return TokenizeWordpiece(input);
+}
+
+WordpieceTokenizerResult BertTokenizer::TokenizeWordpiece(
+    const std::string& input) {
+  WordpieceTokenizerResult result;
+  std::vector<std::string>& subwords = result.subwords;
+  std::vector<int>& wp_absolute_begin_offset = result.wp_begin_offset;
+  std::vector<int>& wp_absolute_end_offset = result.wp_end_offset;
+
+  std::vector<absl::string_view> tokens;
+  std::vector<int64> begin_offsets;
+  std::vector<int64> end_offsets;
+
+  // Run through tokenize function
+  tensorflow::text::RegexSplit(input, delim_re_, true, include_delim_re_,
+                               &tokens, &begin_offsets, &end_offsets);
+
+  for (int token_index = 0; token_index < tokens.size(); token_index++) {
+    auto& token = tokens[token_index];
+    int num_word_pieces = 0;
+    tensorflow::text::LookupStatus status = WordpieceTokenize(
+        token, options_.max_bytes_per_token, options_.max_chars_per_subtoken,
+        options_.suffix_indicator, options_.use_unknown_token,
+        options_.unknown_token, options_.split_unknown_chars, &vocab_,
+        &subwords, &wp_absolute_begin_offset, &wp_absolute_end_offset,
+        &num_word_pieces);
+
+    result.row_lengths.emplace_back(num_word_pieces);
+    // for the last num_word_pieces added into wp_absolute_begin_offset and
+    // wp_absolute_end_offset, offset them with begin_offsets[token_index]
+    int absolute_offset_size = wp_absolute_begin_offset.size();
+    for (int i = num_word_pieces; i > 0; i--) {
+      wp_absolute_begin_offset[absolute_offset_size - i] +=
+          begin_offsets[token_index];
+      wp_absolute_end_offset[absolute_offset_size - i] +=
+          begin_offsets[token_index];
+    }
+    if (!status.success) {
+      return result;
+    }
+  }
+
+  return result;
+}
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h b/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h
new file mode 100644
index 0000000..14a006c
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h
@@ -0,0 +1,149 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_BERT_TOKENIZER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_BERT_TOKENIZER_H_
+
+#include <fstream>
+#include <string>
+#include <vector>
+
+#include "absl/container/flat_hash_map.h"
+#include "re2/re2.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+#include "tensorflow_lite_support/cc/utils/common_utils.h"
+#include "tensorflow_text/core/kernels/regex_split.h"
+#include "tensorflow_text/core/kernels/wordpiece_tokenizer.h"
+
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+constexpr char kDefaultDelimRe[] =
+    R"((\s+|[!-/]|[:-@]|[\[-`]|[{-~]|[\p{P}]|[\x{4E00}-\x{9FFF}]|[\x{3400}-\x{4DBF}]|[\x{20000}-\x{2A6DF}]|[\x{2A700}-\x{2B73F}]|[\x{2B740}-\x{2B81F}]|[\x{2B820}-\x{2CEAF}]|[\x{F900}-\x{FAFF}]|[\x{2F800}-\x{2FA1F}]))";
+constexpr char kDefaultIncludeDelimRe[] =
+    R"(([!-/]|[:-@]|[\[-`]|[{-~]|[\p{P}]|[\x{4E00}-\x{9FFF}]|[\x{3400}-\x{4DBF}]|[\x{20000}-\x{2A6DF}]|[\x{2A700}-\x{2B73F}]|[\x{2B740}-\x{2B81F}]|[\x{2B820}-\x{2CEAF}]|[\x{F900}-\x{FAFF}]|[\x{2F800}-\x{2FA1F}]))";
+constexpr int kDefaultMaxBytesPerToken = 100;
+constexpr int kDefaultMaxCharsPerSubToken = 100;
+constexpr char kDefaultSuffixIndicator[] = "##";
+constexpr bool kDefaultUseUnknownToken = true;
+constexpr char kDefaultUnknownToken[] = "[UNK]";
+constexpr bool kDefaultSplitUnknownChars = false;
+
+// Result of wordpiece tokenization including subwords and offsets.
+// Example:
+// input:                tokenize     me  please
+// subwords:             token ##ize  me  plea ##se
+// wp_begin_offset:     [0,      5,   9,  12,    16]
+// wp_end_offset:       [     5,    8,  11,   16,  18]
+// row_lengths:         [2,          1,  1]
+struct WordpieceTokenizerResult : TokenizerResult {
+  std::vector<int> wp_begin_offset;
+  std::vector<int> wp_end_offset;
+  std::vector<int> row_lengths;
+};
+// Options to create a BertTokenizer.
+struct BertTokenizerOptions {
+  int max_bytes_per_token = kDefaultMaxBytesPerToken;
+  int max_chars_per_subtoken = kDefaultMaxCharsPerSubToken;
+  std::string suffix_indicator = kDefaultSuffixIndicator;
+  bool use_unknown_token = kDefaultUseUnknownToken;
+  std::string unknown_token = kDefaultUnknownToken;
+  bool split_unknown_chars = kDefaultSplitUnknownChars;
+  std::string delim_str = kDefaultDelimRe;
+  std::string include_delim_str = kDefaultIncludeDelimRe;
+};
+
+// A flat-hash-map based implementation of WordpieceVocab, used in
+// BertTokenizer to invoke tensorflow::text::WordpieceTokenize within.
+class FlatHashMapBackedWordpiece : public tensorflow::text::WordpieceVocab {
+ public:
+  explicit FlatHashMapBackedWordpiece(const std::vector<std::string>& vocab);
+
+  tensorflow::text::LookupStatus Contains(absl::string_view key,
+                                          bool* value) const override;
+  bool LookupId(absl::string_view key, int* result) const;
+  bool LookupWord(int vocab_id, absl::string_view* result) const;
+  int VocabularySize() const { return vocab_.size(); }
+
+ private:
+  // All words indexed position in vocabulary file.
+  std::vector<std::string> vocab_;
+  absl::flat_hash_map<absl::string_view, int> index_map_;
+};
+
+// Wordpiece tokenizer for bert models. Initialized with a vocab file or vector.
+class BertTokenizer : public tflite::support::text::tokenizer::Tokenizer {
+ public:
+  // Initialize the tokenizer from vocab vector and tokenizer configs.
+  explicit BertTokenizer(const std::vector<std::string>& vocab,
+                         const BertTokenizerOptions& options = {})
+      : vocab_{FlatHashMapBackedWordpiece(vocab)},
+        options_{options},
+        delim_re_{options.delim_str},
+        include_delim_re_{options.include_delim_str} {}
+
+  // Initialize the tokenizer from file path to vocab and tokenizer configs.
+  explicit BertTokenizer(const std::string& path_to_vocab,
+                         const BertTokenizerOptions& options = {})
+      : BertTokenizer(utils::LoadVocabFromFile(path_to_vocab), options) {}
+
+  // Initialize the tokenizer from buffer and size of vocab and tokenizer
+  // configs.
+  BertTokenizer(const char* vocab_buffer_data, size_t vocab_buffer_size,
+                const BertTokenizerOptions& options = {})
+      : BertTokenizer(
+            utils::LoadVocabFromBuffer(vocab_buffer_data, vocab_buffer_size),
+            options) {}
+
+  // Perform tokenization, return tokenized results containing the subwords.
+  TokenizerResult Tokenize(const std::string& input) override;
+
+  // Perform tokenization, return wordpiece-specific tokenized result including
+  // subwords and offsets
+  WordpieceTokenizerResult TokenizeWordpiece(const std::string& input);
+
+  // Check if a certain key is included in the vocab.
+  tensorflow::text::LookupStatus Contains(const absl::string_view key,
+                                          bool* value) const {
+    return vocab_.Contains(key, value);
+  }
+
+  // Find the id of a wordpiece.
+  bool LookupId(absl::string_view key, int* result) const override {
+    return vocab_.LookupId(key, result);
+  }
+
+  // Find the wordpiece from an id.
+  bool LookupWord(int vocab_id, absl::string_view* result) const override {
+    return vocab_.LookupWord(vocab_id, result);
+  }
+
+  int VocabularySize() const { return vocab_.VocabularySize(); }
+
+ private:
+  tflite::support::text::tokenizer::FlatHashMapBackedWordpiece vocab_;
+  BertTokenizerOptions options_;
+  RE2 delim_re_;
+  RE2 include_delim_re_;
+};
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_BERT_TOKENIZER_H_
diff --git a/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer_jni.cc b/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer_jni.cc
new file mode 100644
index 0000000..442d06e
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer_jni.cc
@@ -0,0 +1,87 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include <string>
+
+#include "absl/memory/memory.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+
+namespace tflite {
+namespace support {
+
+using ::tflite::support::text::tokenizer::BertTokenizer;
+using ::tflite::support::text::tokenizer::BertTokenizerOptions;
+using ::tflite::support::utils::StringListToVector;
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_BertTokenizer_nativeLoadResource(  // NOLINT
+    JNIEnv* env, jobject thiz, jobject vocab_list, jint max_bytes_per_token,
+    jint max_chars_per_sub_token, jstring jsuffix_indicator,
+    jboolean use_unknown_token, jstring junknown_token,
+    jboolean split_unknown_chars) {
+  // Convert java.util.List<String> into std::vector<string>
+  std::vector<std::string> vocab = StringListToVector(env, vocab_list);
+
+  // Convert jstrings to std::string
+  const char* raw_suffix_indicator =
+      env->GetStringUTFChars(jsuffix_indicator, JNI_FALSE);
+  std::string suffix_indicator(raw_suffix_indicator);
+
+  const char* raw_unknown_token =
+      env->GetStringUTFChars(junknown_token, JNI_FALSE);
+  std::string unknown_token(raw_unknown_token);
+
+  auto handle = absl::make_unique<BertTokenizer>(
+      vocab, BertTokenizerOptions{
+                 .max_bytes_per_token = max_bytes_per_token,
+                 .max_chars_per_subtoken = max_chars_per_sub_token,
+                 .suffix_indicator = suffix_indicator,
+                 .use_unknown_token = static_cast<bool>(use_unknown_token),
+                 .unknown_token = unknown_token,
+                 .split_unknown_chars = static_cast<bool>(split_unknown_chars),
+                 .delim_str = text::tokenizer::kDefaultDelimRe,
+                 .include_delim_str = text::tokenizer::kDefaultIncludeDelimRe});
+
+  env->ReleaseStringUTFChars(jsuffix_indicator, raw_suffix_indicator);
+  env->ReleaseStringUTFChars(junknown_token, raw_unknown_token);
+
+  return reinterpret_cast<jlong>(handle.release());
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_BertTokenizer_nativeUnloadResource(  // NOLINT
+    JNIEnv* env, jobject thiz, jlong handle) {
+  delete reinterpret_cast<BertTokenizer*>(handle);
+  return 0;
+}
+
+extern "C" JNIEXPORT jobjectArray JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_BertTokenizer_nativeTokenize(
+    JNIEnv* env, jobject thiz, jlong handle, jstring jtext) {
+  return nativeTokenize(env, handle, jtext);
+}
+
+extern "C" JNIEXPORT jintArray JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_BertTokenizer_nativeConvertTokensToIds(  // NOLINT
+    JNIEnv* env, jobject thiz, jlong handle, jobjectArray jtokens) {
+  return nativeConvertTokensToIds(env, handle, jtokens);
+}
+
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.cc b/tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.cc
new file mode 100644
index 0000000..38aff88
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.cc
@@ -0,0 +1,125 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h"
+
+#include <iostream>
+
+#include "absl/strings/str_cat.h"
+#include "absl/strings/substitute.h"
+#include "tensorflow_lite_support/cc/utils/common_utils.h"
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+namespace {
+constexpr char kStart[] = "<START>";
+constexpr char kPad[] = "<PAD>";
+constexpr char kUnknown[] = "<UNKNOWN>";
+
+void buildIndexTokenMap(
+    const absl::node_hash_map<std::string, int>& token_index_map,
+    absl::node_hash_map<int, absl::string_view>* index_token_map) {
+  for (const auto& token : token_index_map) {
+    (*index_token_map)[token.second] = token.first;
+  }
+}
+
+}  // namespace
+
+// RE2::FindAndConsume requires the delim_re_ to have a matching group in order
+// to capture the matched delimiter length. Surround the regex with a
+// parenthesis to create a matching group, it's fine if the regex is already
+// surrounded by parenthesis.
+RegexTokenizer::RegexTokenizer(const std::string& regex_pattern,
+                               const std::string& path_to_vocab)
+    : delim_re_{absl::Substitute("($0)", regex_pattern)},
+      token_index_map_{utils::LoadVocabAndIndexFromFile(path_to_vocab)} {
+  buildIndexTokenMap(token_index_map_, &index_token_map_);
+}
+
+RegexTokenizer::RegexTokenizer(const std::string& regex_pattern,
+                               const char* vocab_buffer_data,
+                               size_t vocab_buffer_size)
+    : delim_re_{absl::Substitute("($0)", regex_pattern)},
+      token_index_map_{utils::LoadVocabAndIndexFromBuffer(vocab_buffer_data,
+                                                          vocab_buffer_size)} {
+  buildIndexTokenMap(token_index_map_, &index_token_map_);
+}
+
+TokenizerResult RegexTokenizer::Tokenize(const std::string& input) {
+  absl::string_view leftover(input.data());
+  absl::string_view last_end = leftover;
+
+  TokenizerResult result;
+
+  // Keep looking for split points until we have reached the end of the input.
+  absl::string_view extracted_delim_token;
+  while (RE2::FindAndConsume(&leftover, delim_re_, &extracted_delim_token)) {
+    absl::string_view token(last_end.data(),
+                            extracted_delim_token.data() - last_end.data());
+    bool has_non_empty_token = token.length() > 0;
+
+    last_end = leftover;
+
+    // Mark the end of the previous token, only if there was something.
+    if (has_non_empty_token) {
+      result.subwords.push_back(std::string(token));
+    }
+  }
+
+  // Close the last token.
+  if (!leftover.empty()) {
+    result.subwords.push_back(std::string(leftover));
+  }
+
+  return result;
+}
+
+bool RegexTokenizer::LookupId(absl::string_view key, int* result) const {
+  auto it = token_index_map_.find(key);
+  if (it == token_index_map_.end()) {
+    return false;
+  }
+  *result = it->second;
+  return true;
+}
+
+bool RegexTokenizer::LookupWord(int vocab_id, absl::string_view* result) const {
+  auto it = index_token_map_.find(vocab_id);
+  if (it == index_token_map_.end()) {
+    return false;
+  }
+  *result = it->second;
+  return true;
+}
+
+bool RegexTokenizer::GetStartToken(int* start_token) {
+  return LookupId(kStart, start_token);
+}
+
+bool RegexTokenizer::GetPadToken(int* pad_token) {
+  return LookupId(kPad, pad_token);
+}
+
+bool RegexTokenizer::GetUnknownToken(int* unknown_token) {
+  return LookupId(kUnknown, unknown_token);
+}
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h b/tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h
new file mode 100644
index 0000000..c53ae49
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h
@@ -0,0 +1,59 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_REGEX_TOKENIZER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_REGEX_TOKENIZER_H_
+
+#include "absl/container/node_hash_map.h"
+#include "re2/re2.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+// Tokenizer to load a vocabulary and split text by regular expressions.
+class RegexTokenizer : public Tokenizer {
+ public:
+  explicit RegexTokenizer(const std::string& regex_pattern,
+                          const std::string& path_to_vocab);
+
+  explicit RegexTokenizer(const std::string& regex_pattern,
+                          const char* vocab_buffer_data,
+                          size_t vocab_buffer_size);
+
+  TokenizerResult Tokenize(const std::string& input) override;
+
+  bool LookupId(absl::string_view key, int* result) const override;
+
+  bool LookupWord(int vocab_id, absl::string_view* result) const override;
+
+  bool GetStartToken(int* start_token);
+  bool GetPadToken(int* pad_token);
+  bool GetUnknownToken(int* unknown_token);
+
+ private:
+  RE2 delim_re_;
+  absl::node_hash_map<std::string, int> token_index_map_;
+  absl::node_hash_map<int, absl::string_view> index_token_map_;
+};
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_REGEX_TOKENIZER_H_
diff --git a/tensorflow_lite_support/cc/text/tokenizers/sentencepiece_jni.cc b/tensorflow_lite_support/cc/text/tokenizers/sentencepiece_jni.cc
new file mode 100644
index 0000000..88065e2
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/sentencepiece_jni.cc
@@ -0,0 +1,64 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include <cstring>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/memory/memory.h"
+#include "absl/strings/str_split.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+
+namespace tflite {
+namespace support {
+
+using ::tflite::support::text::tokenizer::SentencePieceTokenizer;
+using ::tflite::support::utils::GetMappedFileBuffer;
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_SentencePieceTokenizer_nativeLoadResource(  // NOLINT
+    JNIEnv* env, jobject obj, jobject model_buffer) {
+  auto model = GetMappedFileBuffer(env, model_buffer);
+  auto handle =
+      absl::make_unique<SentencePieceTokenizer>(model.data(), model.size());
+  return reinterpret_cast<jlong>(handle.release());
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_SentencePieceTokenizer_nativeUnloadResource(  // NOLINT
+    JNIEnv* env, jobject obj, jlong handle) {
+  delete reinterpret_cast<SentencePieceTokenizer*>(handle);
+  return 0;
+}
+
+extern "C" JNIEXPORT jobjectArray JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_SentencePieceTokenizer_nativeTokenize(  // NOLINT
+    JNIEnv* env, jobject thiz, jlong handle, jstring jtext) {
+  return nativeTokenize(env, handle, jtext);
+}
+
+extern "C" JNIEXPORT jintArray JNICALL
+Java_org_tensorflow_lite_support_text_tokenizers_SentencePieceTokenizer_nativeConvertTokensToIds(  // NOLINT
+    JNIEnv* env, jobject thiz, jlong handle, jobjectArray jtokens) {
+  return nativeConvertTokensToIds(env, handle, jtokens);
+}
+
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h b/tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h
new file mode 100644
index 0000000..ed5d3da
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h
@@ -0,0 +1,74 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_SENTENCEPIECE_TOKENIZER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_SENTENCEPIECE_TOKENIZER_H_
+
+#include <fstream>
+#include <string>
+#include <vector>
+
+#include "src/sentencepiece_processor.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+// SentencePiece tokenizer. Initialized with a model file.
+class SentencePieceTokenizer : public Tokenizer {
+ public:
+  // Initialize the SentencePiece tokenizer from model file path.
+  explicit SentencePieceTokenizer(const std::string& path_to_model) {
+    CHECK_OK(sp_.Load(path_to_model));
+  }
+
+  explicit SentencePieceTokenizer(const char* spmodel_buffer_data,
+                                  size_t spmodel_buffer_size) {
+    absl::string_view buffer_binary(spmodel_buffer_data, spmodel_buffer_size);
+    CHECK_OK(sp_.LoadFromSerializedProto(buffer_binary));
+  }
+
+  // Perform tokenization, return tokenized results.
+  TokenizerResult Tokenize(const std::string& input) override {
+    TokenizerResult result;
+    std::vector<std::string>& subwords = result.subwords;
+    CHECK_OK(sp_.Encode(input, &subwords));
+    return result;
+  }
+
+  // Find the id of a string token.
+  bool LookupId(absl::string_view key, int* result) const override {
+    *result = sp_.PieceToId(key);
+    return true;
+  }
+
+  // Find the string token of an id.
+  bool LookupWord(int vocab_id, absl::string_view* result) const override {
+    *result = sp_.IdToPiece(vocab_id);
+    return true;
+  }
+
+ private:
+  sentencepiece::SentencePieceProcessor sp_;
+};
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_SENTENCEPIECE_TOKENIZER_H_
diff --git a/tensorflow_lite_support/cc/text/tokenizers/tokenizer.h b/tensorflow_lite_support/cc/text/tokenizers/tokenizer.h
new file mode 100644
index 0000000..c754506
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/tokenizer.h
@@ -0,0 +1,55 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_H_
+
+#include <fstream>
+#include <string>
+#include <vector>
+
+#include "absl/strings/string_view.h"
+
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+struct TokenizerResult {
+  std::vector<std::string> subwords;
+};
+
+// Interface of general tokenizer.
+class Tokenizer {
+ public:
+  // Perform tokenization to get tokenized results.
+  virtual TokenizerResult Tokenize(const std::string& input) = 0;
+
+  // Find the id of a string token.
+  virtual bool LookupId(absl::string_view key, int* result) const = 0;
+
+  // Find the string token from an id.
+  virtual bool LookupWord(int vocab_id, absl::string_view* result) const = 0;
+
+  // Destructor.
+  virtual ~Tokenizer() = default;
+};
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_H_
diff --git a/tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.cc b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.cc
new file mode 100644
index 0000000..a72523b
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.cc
@@ -0,0 +1,86 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.h"
+
+namespace tflite {
+namespace support {
+
+using ::tflite::support::text::tokenizer::Tokenizer;
+using ::tflite::support::text::tokenizer::TokenizerResult;
+using ::tflite::support::utils::CheckNotNull;
+using ::tflite::support::utils::JStringToString;
+using ::tflite::support::utils::kIllegalStateException;
+
+jobjectArray nativeTokenize(JNIEnv* env, jlong handle, jstring jtext) {
+  if (handle == 0) {
+    env->ThrowNew(env->FindClass(kIllegalStateException),
+                  "Vocab not initialized!");
+    return nullptr;
+  }
+
+  Tokenizer* tokenizer = reinterpret_cast<Tokenizer*>(handle);
+
+  // Get the tokenization results.
+  const TokenizerResult tokenize_result =
+      tokenizer->Tokenize(JStringToString(env, jtext));
+  std::vector<std::string> subwords = tokenize_result.subwords;
+
+  jclass string_class = CheckNotNull(env, env->FindClass("java/lang/String"));
+  jobjectArray result = CheckNotNull(
+      env, env->NewObjectArray(subwords.size(), string_class, nullptr));
+
+  for (int i = 0; i < subwords.size(); ++i) {
+    jstring text = CheckNotNull(env, env->NewStringUTF(subwords[i].data()));
+    if (env->ExceptionCheck()) {
+      return nullptr;
+    }
+
+    env->SetObjectArrayElement(result, i, text);
+  }
+
+  return result;
+}
+
+jintArray nativeConvertTokensToIds(JNIEnv* env, jlong handle,
+                                   jobjectArray jtokens) {
+  if (handle == 0) {
+    env->ThrowNew(env->FindClass(kIllegalStateException),
+                  "vocab not initialized!");
+    return nullptr;
+  }
+
+  Tokenizer* tokenizer = reinterpret_cast<Tokenizer*>(handle);
+
+  // Get the token ids.
+  const int count = env->GetArrayLength(jtokens);
+  jintArray result = env->NewIntArray(count);
+  jint* jid_ptr = env->GetIntArrayElements(result, nullptr);
+
+  for (int i = 0; i < count; i++) {
+    auto jstr =
+        reinterpret_cast<jstring>(env->GetObjectArrayElement(jtokens, i));
+    const char* token = env->GetStringUTFChars(jstr, JNI_FALSE);
+    int id;
+    tokenizer->LookupId(token, &id);
+    jid_ptr[i] = id;
+    env->ReleaseStringUTFChars(jstr, token);
+  }
+  env->ReleaseIntArrayElements(result, jid_ptr, 0);
+  return result;
+}
+
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.h b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.h
new file mode 100644
index 0000000..fc7285c
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_jni_lib.h
@@ -0,0 +1,36 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_JNI_LIB_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_JNI_LIB_H_
+
+#include <jni.h>
+
+#include <string>
+
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+
+namespace tflite {
+namespace support {
+
+jobjectArray nativeTokenize(JNIEnv* env, jlong handle, jstring jtext);
+
+jintArray nativeConvertTokensToIds(JNIEnv* env, jlong handle,
+                                   jobjectArray jtokens);
+
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_JNI_LIB_H_
diff --git a/tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.cc b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.cc
new file mode 100644
index 0000000..3e81c47
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.cc
@@ -0,0 +1,136 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.h"
+
+#include "absl/status/status.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+
+using ::tflite::ProcessUnit;
+using ::tflite::SentencePieceTokenizerOptions;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::StatusOr;
+using ::tflite::support::TfLiteSupportStatus;
+
+namespace {
+
+StatusOr<absl::string_view> CheckAndLoadFirstAssociatedFile(
+    const flatbuffers::Vector<flatbuffers::Offset<tflite::AssociatedFile>>*
+        associated_files,
+    const tflite::metadata::ModelMetadataExtractor* metadata_extractor) {
+  if (associated_files == nullptr || associated_files->size() < 1 ||
+      associated_files->Get(0)->name() == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "Invalid vocab_file from input process unit.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+  ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
+                   metadata_extractor->GetAssociatedFile(
+                       associated_files->Get(0)->name()->str()));
+  return vocab_buffer;
+}
+}  // namespace
+
+StatusOr<std::unique_ptr<Tokenizer>> CreateTokenizerFromProcessUnit(
+    const tflite::ProcessUnit* tokenizer_process_unit,
+    const tflite::metadata::ModelMetadataExtractor* metadata_extractor) {
+  if (metadata_extractor == nullptr || tokenizer_process_unit == nullptr) {
+    return CreateStatusWithPayload(
+        absl::StatusCode::kInvalidArgument,
+        "No metadata or input process unit found.",
+        TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+  switch (tokenizer_process_unit->options_type()) {
+    case ProcessUnitOptions_BertTokenizerOptions: {
+      const tflite::BertTokenizerOptions* options =
+          tokenizer_process_unit->options_as<tflite::BertTokenizerOptions>();
+      ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
+                       CheckAndLoadFirstAssociatedFile(options->vocab_file(),
+                                                       metadata_extractor));
+      return absl::make_unique<BertTokenizer>(vocab_buffer.data(),
+                                              vocab_buffer.size());
+    }
+    case ProcessUnitOptions_SentencePieceTokenizerOptions: {
+      const tflite::SentencePieceTokenizerOptions* options =
+          tokenizer_process_unit->options_as<SentencePieceTokenizerOptions>();
+      ASSIGN_OR_RETURN(absl::string_view model_buffer,
+                       CheckAndLoadFirstAssociatedFile(
+                           options->sentencePiece_model(), metadata_extractor));
+      // TODO(b/160647204): Extract sentence piece model vocabulary
+      return absl::make_unique<SentencePieceTokenizer>(model_buffer.data(),
+                                                       model_buffer.size());
+    }
+    case ProcessUnitOptions_RegexTokenizerOptions: {
+      const tflite::RegexTokenizerOptions* options =
+          tokenizer_process_unit->options_as<RegexTokenizerOptions>();
+      ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
+                       CheckAndLoadFirstAssociatedFile(options->vocab_file(),
+                                                       metadata_extractor));
+      if (options->delim_regex_pattern() == nullptr) {
+        return CreateStatusWithPayload(
+            absl::StatusCode::kInvalidArgument,
+            "Invalid delim_regex_pattern from input process unit.",
+            TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+      }
+
+      std::unique_ptr<RegexTokenizer> regex_tokenizer =
+          absl::make_unique<RegexTokenizer>(
+              options->delim_regex_pattern()->str(), vocab_buffer.data(),
+              vocab_buffer.size());
+
+      int unknown_token_id = 0;
+      if (!regex_tokenizer->GetUnknownToken(&unknown_token_id)) {
+        return CreateStatusWithPayload(
+            absl::StatusCode::kInvalidArgument,
+            "RegexTokenizer doesn't have <UNKNOWN> token.",
+            TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+      }
+
+      int pad_token_id = 0;
+      if (!regex_tokenizer->GetPadToken(&pad_token_id)) {
+        return CreateStatusWithPayload(
+            absl::StatusCode::kInvalidArgument,
+            "RegexTokenizer doesn't have <PAD> token.",
+            TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+      }
+
+      return regex_tokenizer;
+    }
+    default:
+      return CreateStatusWithPayload(
+          absl::StatusCode::kNotFound,
+          absl::StrCat("Incorrect options_type:",
+                       tokenizer_process_unit->options_type()),
+          TfLiteSupportStatus::kMetadataInvalidTokenizerError);
+  }
+}
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
+
diff --git a/tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.h b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.h
new file mode 100644
index 0000000..2e50a79
--- /dev/null
+++ b/tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.h
@@ -0,0 +1,41 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_UTILS_H_
+#define THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_UTILS_H_
+
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace text {
+namespace tokenizer {
+
+
+// Create a Tokenizer from model metadata by extracting
+tflite::support::StatusOr<std::unique_ptr<Tokenizer>>
+CreateTokenizerFromProcessUnit(
+    const tflite::ProcessUnit* tokenizer_process_unit,
+    const tflite::metadata::ModelMetadataExtractor* metadata_extractor);
+
+}  // namespace tokenizer
+}  // namespace text
+}  // namespace support
+}  // namespace tflite
+
+#endif  // THIRD_PARTY_TENSORFLOW_LITE_SUPPORT_CC_TEXT_TOKENIZERS_TOKENIZER_UTILS_H_
diff --git a/tensorflow_lite_support/cc/utils/BUILD b/tensorflow_lite_support/cc/utils/BUILD
new file mode 100644
index 0000000..07c832f
--- /dev/null
+++ b/tensorflow_lite_support/cc/utils/BUILD
@@ -0,0 +1,32 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "jni_utils",
+    srcs = [
+        "jni_utils.cc",
+    ],
+    hdrs = [
+        "jni_utils.h",
+    ],
+    deps = [
+        "@com_google_absl//absl/strings",
+        "@org_tensorflow//tensorflow/lite/java/jni",
+    ],
+)
+
+cc_library(
+    name = "common_utils",
+    srcs = [
+        "common_utils.cc",
+    ],
+    hdrs = [
+        "common_utils.h",
+    ],
+    deps = [
+        "@com_google_absl//absl/container:node_hash_map",
+        "@com_google_absl//absl/strings",
+    ],
+)
diff --git a/tensorflow_lite_support/cc/utils/common_utils.cc b/tensorflow_lite_support/cc/utils/common_utils.cc
new file mode 100644
index 0000000..61996f4
--- /dev/null
+++ b/tensorflow_lite_support/cc/utils/common_utils.cc
@@ -0,0 +1,96 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/utils/common_utils.h"
+
+#include <fstream>
+
+#include "absl/strings/str_split.h"
+
+namespace tflite {
+namespace support {
+namespace utils {
+namespace {
+struct membuf : std::streambuf {
+  membuf(char* begin, char* end) { this->setg(begin, begin, end); }
+};
+
+void ReadIStreamLineByLine(
+    std::istream* istream,
+    const std::function<void(std::string)>& line_processor) {
+  std::string str;
+  while (std::getline(*istream, str)) {
+    if (!str.empty()) {
+      line_processor(str);
+    }
+  }
+}
+
+absl::node_hash_map<std::string, int> ReadIStreamLineSplits(
+    std::istream* istream) {
+  absl::node_hash_map<std::string, int> vocab_index_map;
+  std::string str;
+  ReadIStreamLineByLine(istream, [&vocab_index_map](const std::string& str) {
+    std::vector<std::string> v = absl::StrSplit(str, ' ');
+    vocab_index_map[v[0]] = std::stoi(v[1]);
+  });
+  return vocab_index_map;
+}
+
+std::vector<std::string> ReadIStreamByLine(std::istream* istream) {
+  std::vector<std::string> vocab_from_file;
+  std::string str;
+
+  ReadIStreamLineByLine(istream, [&vocab_from_file](const std::string& str) {
+    vocab_from_file.push_back(str);
+  });
+  return vocab_from_file;
+}
+
+}  // namespace
+
+std::vector<std::string> LoadVocabFromFile(const std::string& path_to_vocab) {
+  std::vector<std::string> vocab_from_file;
+  std::ifstream in(path_to_vocab.c_str());
+  return ReadIStreamByLine(&in);
+}
+
+std::vector<std::string> LoadVocabFromBuffer(const char* vocab_buffer_data,
+                                             const size_t vocab_buffer_size) {
+  membuf sbuf(const_cast<char*>(vocab_buffer_data),
+              const_cast<char*>(vocab_buffer_data + vocab_buffer_size));
+  std::istream in(&sbuf);
+  return ReadIStreamByLine(&in);
+}
+
+absl::node_hash_map<std::string, int> LoadVocabAndIndexFromFile(
+    const std::string& path_to_vocab) {
+  absl::node_hash_map<std::string, int> vocab_index_map;
+  std::ifstream in(path_to_vocab.c_str());
+  return ReadIStreamLineSplits(&in);
+}
+
+absl::node_hash_map<std::string, int> LoadVocabAndIndexFromBuffer(
+    const char* vocab_buffer_data, const size_t vocab_buffer_size) {
+  membuf sbuf(const_cast<char*>(vocab_buffer_data),
+              const_cast<char*>(vocab_buffer_data + vocab_buffer_size));
+  absl::node_hash_map<std::string, int> vocab_index_map;
+  std::istream in(&sbuf);
+  return ReadIStreamLineSplits(&in);
+}
+
+}  // namespace utils
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/utils/common_utils.h b/tensorflow_lite_support/cc/utils/common_utils.h
new file mode 100644
index 0000000..3623223
--- /dev/null
+++ b/tensorflow_lite_support/cc/utils/common_utils.h
@@ -0,0 +1,49 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_UTILS_COMMON_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_UTILS_COMMON_UTILS_H_
+
+#include <string>
+#include <vector>
+
+#include "absl/container/node_hash_map.h"
+
+namespace tflite {
+namespace support {
+namespace utils {
+
+// Read a vocab file with one vocabulary on each line, create a vector of
+// strings.
+std::vector<std::string> LoadVocabFromFile(const std::string& path_to_vocab);
+
+// read a vocab buffer with one vocab one each line, create a vector of strings
+std::vector<std::string> LoadVocabFromBuffer(const char* vocab_buffer_data,
+                                             const size_t vocab_buffer_size);
+
+// Read a vocab file with one vocabulary and its corresponding index on each
+// line separated by space, create a map of <vocab, index>.
+absl::node_hash_map<std::string, int> LoadVocabAndIndexFromFile(
+    const std::string& path_to_vocab);
+
+// Read a vocab buffer with one vocabulary and its corresponding index on each
+// line separated by space, create a map of <vocab, index>.
+absl::node_hash_map<std::string, int> LoadVocabAndIndexFromBuffer(
+    const char* vocab_buffer_data, const size_t vocab_buffer_size);
+}  // namespace utils
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_UTILS_COMMON_UTILS_H_
diff --git a/tensorflow_lite_support/cc/utils/jni_utils.cc b/tensorflow_lite_support/cc/utils/jni_utils.cc
new file mode 100644
index 0000000..25cf326
--- /dev/null
+++ b/tensorflow_lite_support/cc/utils/jni_utils.cc
@@ -0,0 +1,100 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+
+#include <string.h>
+
+namespace tflite {
+namespace support {
+namespace utils {
+
+std::string JStringToString(JNIEnv* env, jstring jstr) {
+  if (jstr == nullptr) {
+    return std::string();
+  }
+  const char* cstring = env->GetStringUTFChars(jstr, nullptr);
+  std::string result(cstring);
+  env->ReleaseStringUTFChars(jstr, cstring);
+  return result;
+}
+
+std::vector<std::string> StringListToVector(JNIEnv* env, jobject list_object) {
+  jobject j_iterator = env->CallObjectMethod(
+      list_object, env->GetMethodID(env->GetObjectClass(list_object),
+                                    "iterator", "()Ljava/util/Iterator;"));
+  std::vector<std::string> result;
+  jmethodID has_next =
+      env->GetMethodID(env->GetObjectClass(j_iterator), "hasNext", "()Z");
+  jmethodID get_next = env->GetMethodID(env->GetObjectClass(j_iterator), "next",
+                                        "()Ljava/lang/Object;");
+  while (env->CallBooleanMethod(j_iterator, has_next)) {
+    jstring jstr =
+        static_cast<jstring>(env->CallObjectMethod(j_iterator, get_next));
+    const char* raw_str = env->GetStringUTFChars(jstr, JNI_FALSE);
+    result.emplace_back(std::string(raw_str));
+    env->ReleaseStringUTFChars(jstr, raw_str);
+  }
+  return result;
+}
+
+absl::string_view GetMappedFileBuffer(JNIEnv* env, const jobject& file_buffer) {
+  return absl::string_view(
+      static_cast<char*>(env->GetDirectBufferAddress(file_buffer)),
+      static_cast<size_t>(env->GetDirectBufferCapacity(file_buffer)));
+}
+
+jbyteArray CreateByteArray(JNIEnv* env, const jbyte* data, int num_bytes) {
+  jbyteArray ret = env->NewByteArray(num_bytes);
+  env->SetByteArrayRegion(ret, 0, num_bytes, data);
+
+  return ret;
+}
+
+void ThrowException(JNIEnv* env, const char* clazz, const char* fmt, ...) {
+  va_list args;
+  va_start(args, fmt);
+  const size_t max_msg_len = 512;
+  auto* message = static_cast<char*>(malloc(max_msg_len));
+  if (message && (vsnprintf(message, max_msg_len, fmt, args) >= 0)) {
+    ThrowExceptionWithMessage(env, clazz, message);
+  } else {
+    ThrowExceptionWithMessage(env, clazz, "");
+  }
+  if (message) {
+    free(message);
+  }
+  va_end(args);
+}
+
+void ThrowExceptionWithMessage(JNIEnv* env, const char* clazz,
+                               const char* message) {
+  jclass e_class = env->FindClass(clazz);
+  if (strcmp(clazz, kAssertionError) == 0) {
+    // AssertionError cannot use ThrowNew in Java 7
+    jmethodID constructor =
+        env->GetMethodID(e_class, "<init>", "(Ljava/lang/Object;)V");
+    jstring jstr_message = env->NewStringUTF(message);
+    jobject e_object = env->NewObject(e_class, constructor,
+                                      static_cast<jobject>(jstr_message));
+    env->Throw(static_cast<jthrowable>(e_object));
+    return;
+  }
+  env->ThrowNew(e_class, message);
+}
+
+}  // namespace utils
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/cc/utils/jni_utils.h b/tensorflow_lite_support/cc/utils/jni_utils.h
new file mode 100644
index 0000000..4a4aae4
--- /dev/null
+++ b/tensorflow_lite_support/cc/utils/jni_utils.h
@@ -0,0 +1,91 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CC_UTILS_JNI_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_CC_UTILS_JNI_UTILS_H_
+
+#include <jni.h>
+
+#include <functional>
+#include <string>
+#include <vector>
+
+#include "absl/strings/string_view.h"
+
+namespace tflite {
+namespace support {
+namespace utils {
+
+const char kIllegalArgumentException[] = "java/lang/IllegalArgumentException";
+const char kIllegalStateException[] = "java/lang/IllegalStateException";
+const char kNullPointerException[] = "java/lang/NullPointerException";
+const char kIndexOutOfBoundsException[] = "java/lang/IndexOutOfBoundsException";
+const char kUnsupportedOperationException[] =
+    "java/lang/UnsupportedOperationException";
+const char kAssertionError[] = "java/lang/AssertionError";
+
+constexpr int kInvalidPointer = 0;
+
+// Check if t is nullptr, throw IllegalStateException if it is.
+// Used to verify different types of jobjects are correctly created from jni.
+template <typename T>
+T CheckNotNull(JNIEnv* env, T&& t) {
+  if (t == nullptr) {
+    env->ThrowNew(env->FindClass(kIllegalStateException), "");
+    return nullptr;
+  }
+  return std::forward<T>(t);
+}
+
+// Converts a std::vector<T> into a Java ArrayList using a converter, which
+// processes a single element in the vector before adding it to the ArrayList.
+template <typename T>
+jobject ConvertVectorToArrayList(JNIEnv* env, const std::vector<T>& results,
+                                 std::function<jobject(T)> converter) {
+  jclass array_list_class = env->FindClass("java/util/ArrayList");
+  jmethodID array_list_ctor =
+      env->GetMethodID(array_list_class, "<init>", "(I)V");
+  jint initial_capacity = static_cast<jint>(results.size());
+  jobject array_list_object =
+      env->NewObject(array_list_class, array_list_ctor, initial_capacity);
+  jmethodID array_list_add_method =
+      env->GetMethodID(array_list_class, "add", "(Ljava/lang/Object;)Z");
+
+  for (const auto& ans : results) {
+    env->CallBooleanMethod(array_list_object, array_list_add_method,
+                           converter(ans));
+  }
+  return array_list_object;
+}
+
+std::string JStringToString(JNIEnv* env, jstring jstr);
+
+std::vector<std::string> StringListToVector(JNIEnv* env, jobject list_object);
+
+// Gets a mapped file buffer from a java object representing a file.
+absl::string_view GetMappedFileBuffer(JNIEnv* env, const jobject& file_buffer);
+
+// Creates a Java byte array object based on the input data.
+jbyteArray CreateByteArray(JNIEnv* env, const jbyte* data, int num_bytes);
+
+void ThrowException(JNIEnv* env, const char* clazz, const char* fmt, ...);
+
+void ThrowExceptionWithMessage(JNIEnv* env, const char* clazz,
+                               const char* message);
+
+}  // namespace utils
+}  // namespace support
+}  // namespace tflite
+#endif  // TENSORFLOW_LITE_SUPPORT_CC_UTILS_JNI_UTILS_H_
diff --git a/tensorflow_lite_support/codegen/BUILD b/tensorflow_lite_support/codegen/BUILD
new file mode 100644
index 0000000..b224f98
--- /dev/null
+++ b/tensorflow_lite_support/codegen/BUILD
@@ -0,0 +1,86 @@
+# The tools for generating wrapper classes for a TFLite model with metadata.
+
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "utils",
+    srcs = [
+        "utils.cc",
+    ],
+    hdrs = [
+        "utils.h",
+    ],
+    deps = [
+    ],
+)
+
+cc_library(
+    name = "code_generator",
+    srcs = [
+        "code_generator.cc",
+    ],
+    hdrs = [
+        "code_generator.h",
+    ],
+    deps = [
+        ":utils",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+    ],
+)
+
+cc_library(
+    name = "metadata_helper",
+    srcs = [
+        "metadata_helper.cc",
+    ],
+    hdrs = [
+        "metadata_helper.h",
+    ],
+    deps = [
+        ":utils",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "@org_tensorflow//tensorflow/lite/schema:schema_fbs",
+    ],
+)
+
+cc_library(
+    name = "android_java_generator",
+    srcs = [
+        "android_java_generator.cc",
+    ],
+    hdrs = [
+        "android_java_generator.h",
+    ],
+    deps = [
+        ":code_generator",
+        ":metadata_helper",
+        ":utils",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "@org_tensorflow//tensorflow/lite/schema:schema_fbs",
+    ],
+)
+
+cc_test(
+    name = "code_generator_test",
+    size = "small",
+    srcs = ["code_generator_test.cc"],
+    data = ["//tensorflow_lite_support/metadata:metadata_schema.fbs"],
+    deps = [
+        ":code_generator",
+        "@com_google_googletest//:gtest_main",
+    ],
+)
+
+cc_test(
+    name = "utils_test",
+    srcs = ["utils_test.cc"],
+    deps = [
+        ":utils",
+        "@com_google_googletest//:gtest_main",
+    ],
+)
diff --git a/tensorflow_lite_support/codegen/README.md b/tensorflow_lite_support/codegen/README.md
new file mode 100644
index 0000000..d457edd
--- /dev/null
+++ b/tensorflow_lite_support/codegen/README.md
@@ -0,0 +1,13 @@
+# TensorFlow Lite Android Wrapper Code Generator
+
+For TensorFlow Lite model enhanced with [metadata](https://www.tensorflow.org/lite/convert/metadata.md),
+developers can use the TensorFlow Lite Android wrapper code generator to create
+platform specific wrapper code. The wrapper code removes the need to interact
+directly with `ByteBuffer`. Instead, developers can interact with the TensorFlow
+Lite model with typed objects such as `Bitmap` and `Rect`.
+
+The usefulness of the code generator depend on the completeness of the
+TensorFlow Lite model's metadata entry. Refer to the `<Codegen usage>` section
+under relevant fields in
+[metadata_schema.fbs](https://github.com/tensorflow/tflite-support/blob/master/tensorflow_lite_support/metadata/metadata_schema.fbs),
+to see how the codegen tool parses each field.
diff --git a/tensorflow_lite_support/codegen/android_java_generator.cc b/tensorflow_lite_support/codegen/android_java_generator.cc
new file mode 100644
index 0000000..097119f
--- /dev/null
+++ b/tensorflow_lite_support/codegen/android_java_generator.cc
@@ -0,0 +1,1017 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+// This file contains the logic of android model wrapper generation.
+//
+// At the beginning is the helper functions handling metadata and code writer.
+//
+// Codes are generated in every `Generate{FOO}` functions. Gradle and Manifest
+// files are simple. The wrapper file generation is a bit complex so we divided
+// it into several sub-functions.
+//
+// The structure of the wrapper file looks like:
+//
+// [ imports ]
+// [ class ]
+//   [ inner "Outputs" class ]
+//   [ innner "Metadata" class ]
+//   [ APIs ] ( including ctors, public APIs and private APIs )
+//
+// We tried to mostly write it in a "template-generation" way. `CodeWriter` does
+// the job as a template renderer. To avoid repeatedly setting the token values,
+// helper functions `SetCodeWriterWith{Foo}Info` set the token values with info
+// structures (`TensorInfo` and `ModelInfo`) - the Info structures are
+// intermediate datastructures between Metadata (represented in Flatbuffers) and
+// generated code.
+
+#include "tensorflow_lite_support/codegen/android_java_generator.h"
+
+#include <ctype.h>
+
+#include <algorithm>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "tensorflow_lite_support/codegen/code_generator.h"
+#include "tensorflow_lite_support/codegen/metadata_helper.h"
+#include "tensorflow_lite_support/codegen/utils.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+namespace {
+
+using details_android_java::ModelInfo;
+using details_android_java::TensorInfo;
+
+// Helper class to organize the C++ code block as a generated code block.
+// Using ctor and dtor to simulate an enter/exit schema like `with` in Python.
+class AsBlock {
+ public:
+  AsBlock(CodeWriter* code_writer, const std::string& before,
+          bool trailing_blank_line = false)
+      : code_writer_(code_writer), trailing_blank_line_(trailing_blank_line) {
+    code_writer_->AppendNoNewLine(before);
+    code_writer_->Append(" {");
+    code_writer_->Indent();
+  }
+  ~AsBlock() {
+    code_writer_->Outdent();
+    code_writer_->Append("}");
+    if (trailing_blank_line_) {
+      code_writer_->NewLine();
+    }
+  }
+
+ private:
+  CodeWriter* code_writer_;
+  bool trailing_blank_line_;
+};
+
+// Declare the functions first, so that the functions can follow a logical
+// order.
+bool GenerateWrapperClass(CodeWriter*, const ModelInfo&, ErrorReporter*);
+bool GenerateWrapperImports(CodeWriter*, const ModelInfo&, ErrorReporter*);
+bool GenerateWrapperInputs(CodeWriter*, const ModelInfo&, ErrorReporter*);
+bool GenerateWrapperOutputs(CodeWriter*, const ModelInfo&, ErrorReporter*);
+bool GenerateWrapperMetadata(CodeWriter*, const ModelInfo&, ErrorReporter*);
+bool GenerateWrapperAPI(CodeWriter*, const ModelInfo&, ErrorReporter*);
+
+std::string GetModelVersionedName(const ModelMetadata* metadata) {
+  std::string model_name = "MyModel";
+  if (metadata->name() != nullptr && !(metadata->name()->str().empty())) {
+    model_name = metadata->name()->str();
+  }
+  std::string model_version = "unknown";
+  if (metadata->version() != nullptr && !(metadata->version()->str().empty())) {
+    model_version = metadata->version()->str();
+  }
+  return model_name + " (Version: " + model_version + ")";
+}
+
+TensorInfo CreateTensorInfo(const TensorMetadata* metadata,
+                            const std::string& name, bool is_input, int index,
+                            ErrorReporter* err) {
+  TensorInfo tensor_info;
+  std::string tensor_identifier = is_input ? "input" : "output";
+  tensor_identifier += " " + std::to_string(index);
+  tensor_info.associated_axis_label_index = FindAssociatedFile(
+      metadata, AssociatedFileType_TENSOR_AXIS_LABELS, tensor_identifier, err);
+  tensor_info.associated_value_label_index = FindAssociatedFile(
+      metadata, AssociatedFileType_TENSOR_VALUE_LABELS, tensor_identifier, err);
+  if (is_input && (tensor_info.associated_axis_label_index >= 0 ||
+                   tensor_info.associated_value_label_index >= 0)) {
+    err->Warning(
+        "Found label file on input tensor (%s). Label file for input "
+        "tensor is not supported yet. The "
+        "file will be ignored.",
+        tensor_identifier.c_str());
+  }
+  if (tensor_info.associated_axis_label_index >= 0 &&
+      tensor_info.associated_value_label_index >= 0) {
+    err->Warning(
+        "Found both axis label file and value label file for tensor (%s), "
+        "which is not supported. Only the axis label file will be used.",
+        tensor_identifier.c_str());
+  }
+  tensor_info.is_input = is_input;
+  tensor_info.name = SnakeCaseToCamelCase(name);
+  tensor_info.upper_camel_name = tensor_info.name;
+  tensor_info.upper_camel_name[0] = toupper(tensor_info.upper_camel_name[0]);
+  tensor_info.normalization_unit =
+      FindNormalizationUnit(metadata, tensor_identifier, err);
+  if (metadata->content() != nullptr &&
+      metadata->content()->content_properties() != nullptr) {
+    // Enter tensor wrapper type inferring
+    if (metadata->content()->content_properties_type() ==
+        ContentProperties_ImageProperties) {
+      if (metadata->content()
+              ->content_properties_as_ImageProperties()
+              ->color_space() == ColorSpaceType_RGB) {
+        tensor_info.content_type = "image";
+        tensor_info.wrapper_type = "TensorImage";
+        tensor_info.processor_type = "ImageProcessor";
+        return tensor_info;
+      } else {
+        err->Warning(
+            "Found Non-RGB image on tensor (%s). Codegen currently does not "
+            "support it, and regard it as a plain numeric tensor.",
+            tensor_identifier.c_str());
+      }
+    }
+  }
+  tensor_info.content_type = "tensor";
+  tensor_info.wrapper_type = "TensorBuffer";
+  tensor_info.processor_type = "TensorProcessor";
+  return tensor_info;
+}
+
+ModelInfo CreateModelInfo(const ModelMetadata* metadata,
+                          const std::string& package_name,
+                          const std::string& model_class_name,
+                          const std::string& model_asset_path,
+                          ErrorReporter* err) {
+  ModelInfo model_info;
+  if (!CodeGenerator::VerifyMetadata(metadata, err)) {
+    // TODO(b/150116380): Create dummy model info.
+    err->Error("Validating metadata failed.");
+    return model_info;
+  }
+  model_info.package_name = package_name;
+  model_info.model_class_name = model_class_name;
+  model_info.model_asset_path = model_asset_path;
+  model_info.model_versioned_name = GetModelVersionedName(metadata);
+  const auto* graph = metadata->subgraph_metadata()->Get(0);
+  auto names = CodeGenerator::NameInputsAndOutputs(
+      graph->input_tensor_metadata(), graph->output_tensor_metadata());
+  std::vector<std::string> input_tensor_names = std::move(names.first);
+  std::vector<std::string> output_tensor_names = std::move(names.second);
+
+  for (int i = 0; i < input_tensor_names.size(); i++) {
+    model_info.inputs.push_back(
+        CreateTensorInfo(graph->input_tensor_metadata()->Get(i),
+                         input_tensor_names[i], true, i, err));
+    if (i < input_tensor_names.size() - 1) {
+      model_info.inputs_list += ", ";
+      model_info.input_type_param_list += ", ";
+    }
+    model_info.inputs_list += model_info.inputs[i].name;
+    model_info.input_type_param_list +=
+        model_info.inputs[i].wrapper_type + " " + model_info.inputs[i].name;
+  }
+  for (int i = 0; i < output_tensor_names.size(); i++) {
+    model_info.outputs.push_back(
+        CreateTensorInfo(graph->output_tensor_metadata()->Get(i),
+                         output_tensor_names[i], false, i, err));
+    if (i < output_tensor_names.size() - 1) {
+      model_info.postprocessor_type_param_list += ", ";
+      model_info.postprocessors_list += ", ";
+    }
+    model_info.postprocessors_list +=
+        model_info.outputs[i].name + "Postprocessor";
+    model_info.postprocessor_type_param_list +=
+        model_info.outputs[i].processor_type + " " +
+        model_info.outputs[i].name + "Postprocessor";
+  }
+  return model_info;
+}
+
+void SetCodeWriterWithTensorInfo(CodeWriter* code_writer,
+                                 const TensorInfo& tensor_info) {
+  code_writer->SetTokenValue("NAME", tensor_info.name);
+  code_writer->SetTokenValue("NAME_U", tensor_info.upper_camel_name);
+  code_writer->SetTokenValue("CONTENT_TYPE", tensor_info.content_type);
+  code_writer->SetTokenValue("WRAPPER_TYPE", tensor_info.wrapper_type);
+  std::string wrapper_name = tensor_info.wrapper_type;
+  wrapper_name[0] = tolower(wrapper_name[0]);
+  code_writer->SetTokenValue("WRAPPER_NAME", wrapper_name);
+  code_writer->SetTokenValue("PROCESSOR_TYPE", tensor_info.processor_type);
+  code_writer->SetTokenValue("NORMALIZATION_UNIT",
+                             std::to_string(tensor_info.normalization_unit));
+  code_writer->SetTokenValue(
+      "ASSOCIATED_AXIS_LABEL_INDEX",
+      std::to_string(tensor_info.associated_axis_label_index));
+  code_writer->SetTokenValue(
+      "ASSOCIATED_VALUE_LABEL_INDEX",
+      std::to_string(tensor_info.associated_value_label_index));
+}
+
+void SetCodeWriterWithModelInfo(CodeWriter* code_writer,
+                                const ModelInfo& model_info) {
+  code_writer->SetTokenValue("PACKAGE", model_info.package_name);
+  code_writer->SetTokenValue("MODEL_PATH", model_info.model_asset_path);
+  code_writer->SetTokenValue("MODEL_CLASS_NAME", model_info.model_class_name);
+  // Extra info, half generated.
+  code_writer->SetTokenValue("INPUT_TYPE_PARAM_LIST",
+                             model_info.input_type_param_list);
+  code_writer->SetTokenValue("INPUTS_LIST", model_info.inputs_list);
+  code_writer->SetTokenValue("POSTPROCESSORS_LIST",
+                             model_info.postprocessors_list);
+  code_writer->SetTokenValue("POSTPROCESSOR_TYPE_PARAM_LIST",
+                             model_info.postprocessor_type_param_list);
+}
+
+constexpr char JAVA_DEFAULT_PACKAGE[] = "default";
+
+std::string ConvertPackageToPath(const std::string& package) {
+  if (package == JAVA_DEFAULT_PACKAGE) {
+    return "";
+  }
+  std::string path = package;
+  std::replace(path.begin(), path.end(), '.', '/');
+  return path;
+}
+
+bool IsImageUsed(const ModelInfo& model) {
+  for (const auto& input : model.inputs) {
+    if (input.content_type == "image") {
+      return true;
+    }
+  }
+  for (const auto& output : model.outputs) {
+    if (output.content_type == "image") {
+      return true;
+    }
+  }
+  return false;
+}
+
+// The following functions generates the wrapper Java code for a model.
+
+bool GenerateWrapperFileContent(CodeWriter* code_writer, const ModelInfo& model,
+                                ErrorReporter* err) {
+  code_writer->Append("// Generated by TFLite Support.");
+  code_writer->Append("package {{PACKAGE}};");
+  code_writer->NewLine();
+
+  if (!GenerateWrapperImports(code_writer, model, err)) {
+    err->Error("Fail to generate imports for wrapper class.");
+    return false;
+  }
+  if (!GenerateWrapperClass(code_writer, model, err)) {
+    err->Error("Fail to generate wrapper class.");
+    return false;
+  }
+  code_writer->NewLine();
+  return true;
+}
+
+bool GenerateWrapperImports(CodeWriter* code_writer, const ModelInfo& model,
+                            ErrorReporter* err) {
+  const std::string support_pkg = "org.tensorflow.lite.support.";
+  std::vector<std::string> imports{
+      "android.content.Context",
+      "java.io.IOException",
+      "java.nio.ByteBuffer",
+      "java.nio.FloatBuffer",
+      "java.util.Arrays",
+      "java.util.HashMap",
+      "java.util.List",
+      "java.util.Map",
+      "org.tensorflow.lite.DataType",
+      "org.tensorflow.lite.Tensor",
+      "org.tensorflow.lite.Tensor.QuantizationParams",
+      support_pkg + "common.FileUtil",
+      support_pkg + "common.TensorProcessor",
+      support_pkg + "common.ops.CastOp",
+      support_pkg + "common.ops.DequantizeOp",
+      support_pkg + "common.ops.NormalizeOp",
+      support_pkg + "common.ops.QuantizeOp",
+      support_pkg + "label.Category",
+      support_pkg + "label.TensorLabel",
+      support_pkg + "metadata.MetadataExtractor",
+      support_pkg + "metadata.schema.NormalizationOptions",
+      support_pkg + "model.Model",
+      support_pkg + "tensorbuffer.TensorBuffer",
+  };
+  if (IsImageUsed(model)) {
+    for (const auto& target :
+         {"image.ImageProcessor", "image.TensorImage", "image.ops.ResizeOp",
+          "image.ops.ResizeOp.ResizeMethod"}) {
+      imports.push_back(support_pkg + target);
+    }
+  }
+
+  std::sort(imports.begin(), imports.end());
+  for (const auto& target : imports) {
+    code_writer->SetTokenValue("TARGET", target);
+    code_writer->Append("import {{TARGET}};");
+  }
+  code_writer->NewLine();
+  return true;
+}
+
+bool GenerateWrapperClass(CodeWriter* code_writer, const ModelInfo& model,
+                          ErrorReporter* err) {
+  code_writer->SetTokenValue("MODEL_VERSIONED_NAME",
+                             model.model_versioned_name);
+  code_writer->Append(
+      R"(/** Wrapper class of model {{MODEL_VERSIONED_NAME}} */)");
+  const auto code_block =
+      AsBlock(code_writer, "public class {{MODEL_CLASS_NAME}}");
+  code_writer->Append(R"(private final Metadata metadata;
+private final Model model;
+private static final String MODEL_NAME = "{{MODEL_PATH}}";)");
+  for (const auto& tensor : model.inputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append("private {{PROCESSOR_TYPE}} {{NAME}}Preprocessor;");
+  }
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append("private {{PROCESSOR_TYPE}} {{NAME}}Postprocessor;");
+  }
+  code_writer->NewLine();
+  if (!GenerateWrapperOutputs(code_writer, model, err)) {
+    err->Error("Failed to generate output classes");
+    return false;
+  }
+  code_writer->NewLine();
+  if (!GenerateWrapperMetadata(code_writer, model, err)) {
+    err->Error("Failed to generate the metadata class");
+    return false;
+  }
+  code_writer->NewLine();
+  if (!GenerateWrapperAPI(code_writer, model, err)) {
+    err->Error("Failed to generate the common APIs");
+    return false;
+  }
+  return true;
+}
+
+bool GenerateWrapperOutputs(CodeWriter* code_writer, const ModelInfo& model,
+                            ErrorReporter* err) {
+  code_writer->Append("/** Output wrapper of {@link {{MODEL_CLASS_NAME}}} */");
+  auto class_block = AsBlock(code_writer, "public static class Outputs");
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append("private final {{WRAPPER_TYPE}} {{NAME}};");
+    if (tensor.associated_axis_label_index >= 0) {
+      code_writer->Append("private final List<String> {{NAME}}Labels;");
+    }
+    code_writer->Append(
+        "private final {{PROCESSOR_TYPE}} {{NAME}}Postprocessor;");
+  }
+  // Getters
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->NewLine();
+    if (tensor.associated_axis_label_index >= 0) {
+      if (tensor.content_type == "tensor") {
+        code_writer->Append(
+            R"(public List<Category> get{{NAME_U}}AsCategoryList() {
+  return new TensorLabel({{NAME}}Labels, postprocess{{NAME_U}}({{NAME}})).getCategoryList();
+})");
+      } else {  // image
+        err->Warning(
+            "Axis label for images is not supported. The labels will "
+            "be ignored.");
+      }
+    } else {  // no label
+      code_writer->Append(
+          R"(public {{WRAPPER_TYPE}} get{{NAME_U}}As{{WRAPPER_TYPE}}() {
+  return postprocess{{NAME_U}}({{NAME}});
+})");
+    }
+  }
+  code_writer->NewLine();
+  {
+    const auto ctor_block = AsBlock(
+        code_writer,
+        "Outputs(Metadata metadata, {{POSTPROCESSOR_TYPE_PARAM_LIST}})");
+    for (const auto& tensor : model.outputs) {
+      SetCodeWriterWithTensorInfo(code_writer, tensor);
+      if (tensor.content_type == "image") {
+        code_writer->Append(
+            R"({{NAME}} = new TensorImage(metadata.get{{NAME_U}}Type());
+{{NAME}}.load(TensorBuffer.createFixedSize(metadata.get{{NAME_U}}Shape(), metadata.get{{NAME_U}}Type()));)");
+      } else {  // FEATURE, UNKNOWN
+        code_writer->Append(
+            "{{NAME}} = "
+            "TensorBuffer.createFixedSize(metadata.get{{NAME_U}}Shape(), "
+            "metadata.get{{NAME_U}}Type());");
+      }
+      if (tensor.associated_axis_label_index >= 0) {
+        code_writer->Append("{{NAME}}Labels = metadata.get{{NAME_U}}Labels();");
+      }
+      code_writer->Append(
+          "this.{{NAME}}Postprocessor = {{NAME}}Postprocessor;");
+    }
+  }
+  code_writer->NewLine();
+  {
+    const auto get_buffer_block =
+        AsBlock(code_writer, "Map<Integer, Object> getBuffer()");
+    code_writer->Append("Map<Integer, Object> outputs = new HashMap<>();");
+    for (int i = 0; i < model.outputs.size(); i++) {
+      SetCodeWriterWithTensorInfo(code_writer, model.outputs[i]);
+      code_writer->SetTokenValue("ID", std::to_string(i));
+      code_writer->Append("outputs.put({{ID}}, {{NAME}}.getBuffer());");
+    }
+    code_writer->Append("return outputs;");
+  }
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->NewLine();
+    {
+      auto processor_block =
+          AsBlock(code_writer,
+                  "private {{WRAPPER_TYPE}} "
+                  "postprocess{{NAME_U}}({{WRAPPER_TYPE}} {{WRAPPER_NAME}})");
+      code_writer->Append(
+          "return {{NAME}}Postprocessor.process({{WRAPPER_NAME}});");
+    }
+  }
+  return true;
+}
+
+bool GenerateWrapperMetadata(CodeWriter* code_writer, const ModelInfo& model,
+                             ErrorReporter* err) {
+  code_writer->Append(
+      "/** Metadata accessors of {@link {{MODEL_CLASS_NAME}}} */");
+  const auto class_block = AsBlock(code_writer, "public static class Metadata");
+  for (const auto& tensor : model.inputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(R"(private final int[] {{NAME}}Shape;
+private final DataType {{NAME}}DataType;
+private final QuantizationParams {{NAME}}QuantizationParams;)");
+    if (tensor.normalization_unit >= 0) {
+      code_writer->Append(R"(private final float[] {{NAME}}Mean;
+private final float[] {{NAME}}Stddev;)");
+    }
+  }
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(R"(private final int[] {{NAME}}Shape;
+private final DataType {{NAME}}DataType;
+private final QuantizationParams {{NAME}}QuantizationParams;)");
+    if (tensor.normalization_unit >= 0) {
+      code_writer->Append(R"(private final float[] {{NAME}}Mean;
+private final float[] {{NAME}}Stddev;)");
+    }
+    if (tensor.associated_axis_label_index >= 0 ||
+        tensor.associated_value_label_index >= 0) {
+      code_writer->Append("private final List<String> {{NAME}}Labels;");
+    }
+  }
+  code_writer->NewLine();
+  {
+    const auto ctor_block = AsBlock(
+        code_writer,
+        "public Metadata(ByteBuffer buffer, Model model) throws IOException");
+    code_writer->Append(
+        "MetadataExtractor extractor = new MetadataExtractor(buffer);");
+    for (int i = 0; i < model.inputs.size(); i++) {
+      SetCodeWriterWithTensorInfo(code_writer, model.inputs[i]);
+      code_writer->SetTokenValue("ID", std::to_string(i));
+      code_writer->Append(
+          R"(Tensor {{NAME}}Tensor = model.getInputTensor({{ID}});
+{{NAME}}Shape = {{NAME}}Tensor.shape();
+{{NAME}}DataType = {{NAME}}Tensor.dataType();
+{{NAME}}QuantizationParams = {{NAME}}Tensor.quantizationParams();)");
+      if (model.inputs[i].normalization_unit >= 0) {
+        code_writer->Append(
+            R"(NormalizationOptions {{NAME}}NormalizationOptions =
+    (NormalizationOptions) extractor.getInputTensorMetadata({{ID}}).processUnits({{NORMALIZATION_UNIT}}).options(new NormalizationOptions());
+FloatBuffer {{NAME}}MeanBuffer = {{NAME}}NormalizationOptions.meanAsByteBuffer().asFloatBuffer();
+{{NAME}}Mean = new float[{{NAME}}MeanBuffer.limit()];
+{{NAME}}MeanBuffer.get({{NAME}}Mean);
+FloatBuffer {{NAME}}StddevBuffer = {{NAME}}NormalizationOptions.stdAsByteBuffer().asFloatBuffer();
+{{NAME}}Stddev = new float[{{NAME}}StddevBuffer.limit()];
+{{NAME}}StddevBuffer.get({{NAME}}Stddev);)");
+      }
+    }
+    for (int i = 0; i < model.outputs.size(); i++) {
+      SetCodeWriterWithTensorInfo(code_writer, model.outputs[i]);
+      code_writer->SetTokenValue("ID", std::to_string(i));
+      code_writer->Append(
+          R"(Tensor {{NAME}}Tensor = model.getOutputTensor({{ID}});
+{{NAME}}Shape = {{NAME}}Tensor.shape();
+{{NAME}}DataType = {{NAME}}Tensor.dataType();
+{{NAME}}QuantizationParams = {{NAME}}Tensor.quantizationParams();)");
+      if (model.outputs[i].normalization_unit >= 0) {
+        code_writer->Append(
+            R"(NormalizationOptions {{NAME}}NormalizationOptions =
+    (NormalizationOptions) extractor.getInputTensorMetadata({{ID}}).processUnits({{NORMALIZATION_UNIT}}).options(new NormalizationOptions());
+FloatBuffer {{NAME}}MeanBuffer = {{NAME}}NormalizationOptions.meanAsByteBuffer().asFloatBuffer();
+{{NAME}}Mean = new float[{{NAME}}MeanBuffer.limit()];
+{{NAME}}MeanBuffer.get({{NAME}}Mean);
+FloatBuffer {{NAME}}StddevBuffer = {{NAME}}NormalizationOptions.stdAsByteBuffer().asFloatBuffer();
+{{NAME}}Stddev = new float[{{NAME}}StddevBuffer.limit()];
+{{NAME}}StddevBuffer.get({{NAME}}Stddev);)");
+      }
+      if (model.outputs[i].associated_axis_label_index >= 0) {
+        code_writer->Append(R"(String {{NAME}}LabelsFileName =
+    extractor.getOutputTensorMetadata({{ID}}).associatedFiles({{ASSOCIATED_AXIS_LABEL_INDEX}}).name();
+{{NAME}}Labels = FileUtil.loadLabels(extractor.getAssociatedFile({{NAME}}LabelsFileName));)");
+      } else if (model.outputs[i].associated_value_label_index >= 0) {
+        code_writer->Append(R"(String {{NAME}}LabelsFileName =
+    extractor.getOutputTensorMetadata({{ID}}).associatedFiles({{ASSOCIATED_VALUE_LABEL_INDEX}}).name();
+{{NAME}}Labels = FileUtil.loadLabels(extractor.getAssociatedFile({{NAME}}LabelsFileName));)");
+      }
+    }
+  }
+  for (const auto& tensor : model.inputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(R"(
+public int[] get{{NAME_U}}Shape() {
+  return Arrays.copyOf({{NAME}}Shape, {{NAME}}Shape.length);
+}
+
+public DataType get{{NAME_U}}Type() {
+  return {{NAME}}DataType;
+}
+
+public QuantizationParams get{{NAME_U}}QuantizationParams() {
+  return {{NAME}}QuantizationParams;
+})");
+    if (tensor.normalization_unit >= 0) {
+      code_writer->Append(R"(
+public float[] get{{NAME_U}}Mean() {
+  return Arrays.copyOf({{NAME}}Mean, {{NAME}}Mean.length);
+}
+
+public float[] get{{NAME_U}}Stddev() {
+  return Arrays.copyOf({{NAME}}Stddev, {{NAME}}Stddev.length);
+})");
+    }
+  }
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(R"(
+public int[] get{{NAME_U}}Shape() {
+  return Arrays.copyOf({{NAME}}Shape, {{NAME}}Shape.length);
+}
+
+public DataType get{{NAME_U}}Type() {
+  return {{NAME}}DataType;
+}
+
+public QuantizationParams get{{NAME_U}}QuantizationParams() {
+  return {{NAME}}QuantizationParams;
+})");
+    if (tensor.normalization_unit >= 0) {
+      code_writer->Append(R"(
+public float[] get{{NAME_U}}Mean() {
+  return Arrays.copyOf({{NAME}}Mean, {{NAME}}Mean.length);
+}
+
+public float[] get{{NAME_U}}Stddev() {
+  return Arrays.copyOf({{NAME}}Stddev, {{NAME}}Stddev.length);
+})");
+    }
+    if (tensor.associated_axis_label_index >= 0 ||
+        tensor.associated_value_label_index >= 0) {
+      code_writer->Append(R"(
+public List<String> get{{NAME_U}}Labels() {
+  return {{NAME}}Labels;
+})");
+    }
+  }
+  return true;
+}
+
+bool GenerateWrapperAPI(CodeWriter* code_writer, const ModelInfo& model,
+                        ErrorReporter* err) {
+  code_writer->Append(R"(public Metadata getMetadata() {
+  return metadata;
+}
+)");
+  code_writer->Append(R"(/**
+ * Creates interpreter and loads associated files if needed.
+ *
+ * @throws IOException if an I/O error occurs when loading the tflite model.
+ */
+public static {{MODEL_CLASS_NAME}} newInstance(Context context) throws IOException {
+  return newInstance(context, MODEL_NAME, new Model.Options.Builder().build());
+}
+
+/**
+ * Creates interpreter and loads associated files if needed, but loading another model in the same
+ * input / output structure with the original one.
+ *
+ * @throws IOException if an I/O error occurs when loading the tflite model.
+ */
+public static {{MODEL_CLASS_NAME}} newInstance(Context context, String modelPath) throws IOException {
+  return newInstance(context, modelPath, new Model.Options.Builder().build());
+}
+
+/**
+ * Creates interpreter and loads associated files if needed, with running options configured.
+ *
+ * @throws IOException if an I/O error occurs when loading the tflite model.
+ */
+public static {{MODEL_CLASS_NAME}} newInstance(Context context, Model.Options runningOptions) throws IOException {
+  return newInstance(context, MODEL_NAME, runningOptions);
+}
+
+/**
+ * Creates interpreter for a user-specified model.
+ *
+ * @throws IOException if an I/O error occurs when loading the tflite model.
+ */
+public static {{MODEL_CLASS_NAME}} newInstance(Context context, String modelPath, Model.Options runningOptions) throws IOException {
+  Model model = Model.createModel(context, modelPath, runningOptions);
+  Metadata metadata = new Metadata(model.getData(), model);
+  MyImageClassifier instance = new MyImageClassifier(model, metadata);)");
+  for (const auto& tensor : model.inputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(
+        R"(  instance.reset{{NAME_U}}Preprocessor(
+      instance.buildDefault{{NAME_U}}Preprocessor());)");
+  }
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(
+        R"(  instance.reset{{NAME_U}}Postprocessor(
+      instance.buildDefault{{NAME_U}}Postprocessor());)");
+  }
+  code_writer->Append(R"(  return instance;
+}
+)");
+
+  // Pre, post processor setters
+  for (const auto& tensor : model.inputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(R"(
+public void reset{{NAME_U}}Preprocessor({{PROCESSOR_TYPE}} processor) {
+  {{NAME}}Preprocessor = processor;
+})");
+  }
+  for (const auto& tensor : model.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    code_writer->Append(R"(
+public void reset{{NAME_U}}Postprocessor({{PROCESSOR_TYPE}} processor) {
+  {{NAME}}Postprocessor = processor;
+})");
+  }
+  // Process method
+  code_writer->Append(R"(
+/** Triggers the model. */
+public Outputs process({{INPUT_TYPE_PARAM_LIST}}) {
+  Outputs outputs = new Outputs(metadata, {{POSTPROCESSORS_LIST}});
+  Object[] inputBuffers = preprocessInputs({{INPUTS_LIST}});
+  model.run(inputBuffers, outputs.getBuffer());
+  return outputs;
+}
+
+/** Closes the model. */
+public void close() {
+  model.close();
+}
+)");
+  {
+    auto block =
+        AsBlock(code_writer,
+                "private {{MODEL_CLASS_NAME}}(Model model, Metadata metadata)");
+    code_writer->Append(R"(this.model = model;
+this.metadata = metadata;)");
+  }
+  for (const auto& tensor : model.inputs) {
+    code_writer->NewLine();
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    auto block = AsBlock(
+        code_writer,
+        "private {{PROCESSOR_TYPE}} buildDefault{{NAME_U}}Preprocessor()");
+    code_writer->Append(
+        "{{PROCESSOR_TYPE}}.Builder builder = new "
+        "{{PROCESSOR_TYPE}}.Builder()");
+    if (tensor.content_type == "image") {
+      code_writer->Append(R"(    .add(new ResizeOp(
+        metadata.get{{NAME_U}}Shape()[1],
+        metadata.get{{NAME_U}}Shape()[2],
+        ResizeMethod.NEAREST_NEIGHBOR)))");
+    }
+    if (tensor.normalization_unit >= 0) {
+      code_writer->Append(
+          R"(    .add(new NormalizeOp(metadata.get{{NAME_U}}Mean(), metadata.get{{NAME_U}}Stddev())))");
+    }
+    code_writer->Append(
+        R"(    .add(new QuantizeOp(
+        metadata.get{{NAME_U}}QuantizationParams().getZeroPoint(),
+        metadata.get{{NAME_U}}QuantizationParams().getScale()))
+    .add(new CastOp(metadata.get{{NAME_U}}Type()));
+return builder.build();)");
+  }
+  for (const auto& tensor : model.outputs) {
+    code_writer->NewLine();
+    SetCodeWriterWithTensorInfo(code_writer, tensor);
+    auto block = AsBlock(
+        code_writer,
+        "private {{PROCESSOR_TYPE}} buildDefault{{NAME_U}}Postprocessor()");
+    code_writer->AppendNoNewLine(
+        R"({{PROCESSOR_TYPE}}.Builder builder = new {{PROCESSOR_TYPE}}.Builder()
+    .add(new DequantizeOp(
+        metadata.get{{NAME_U}}QuantizationParams().getZeroPoint(),
+        metadata.get{{NAME_U}}QuantizationParams().getScale())))");
+    if (tensor.normalization_unit >= 0) {
+      code_writer->AppendNoNewLine(R"(
+    .add(new NormalizeOp(metadata.get{{NAME_U}}Mean(), metadata.get{{NAME_U}}Stddev())))");
+    }
+    code_writer->Append(R"(;
+return builder.build();)");
+  }
+  code_writer->NewLine();
+  {
+    const auto block =
+        AsBlock(code_writer,
+                "private Object[] preprocessInputs({{INPUT_TYPE_PARAM_LIST}})");
+    CodeWriter param_list_gen(err);
+    for (const auto& tensor : model.inputs) {
+      SetCodeWriterWithTensorInfo(code_writer, tensor);
+      code_writer->Append("{{NAME}} = {{NAME}}Preprocessor.process({{NAME}});");
+      SetCodeWriterWithTensorInfo(&param_list_gen, tensor);
+      param_list_gen.AppendNoNewLine("{{NAME}}.getBuffer(), ");
+    }
+    param_list_gen.Backspace(2);
+    code_writer->AppendNoNewLine("return new Object[] {");
+    code_writer->AppendNoNewLine(param_list_gen.ToString());
+    code_writer->Append("};");
+  }
+  return true;
+}
+
+bool GenerateBuildGradleContent(CodeWriter* code_writer,
+                                const ModelInfo& model_info) {
+  code_writer->Append(R"(buildscript {
+    repositories {
+        google()
+        jcenter()
+    }
+    dependencies {
+        classpath 'com.android.tools.build:gradle:3.2.1'
+    }
+}
+
+allprojects {
+    repositories {
+        google()
+        jcenter()
+        flatDir {
+            dirs 'libs'
+        }
+    }
+}
+
+apply plugin: 'com.android.library'
+
+android {
+    compileSdkVersion 29
+    defaultConfig {
+        targetSdkVersion 29
+        versionCode 1
+        versionName "1.0"
+    }
+    aaptOptions {
+        noCompress "tflite"
+    }
+    compileOptions {
+        sourceCompatibility = '1.8'
+        targetCompatibility = '1.8'
+    }
+    lintOptions {
+        abortOnError false
+    }
+}
+
+configurations {
+    libMetadata
+}
+
+dependencies {
+    libMetadata 'org.tensorflow:tensorflow-lite-support:0.0.0-experimental-metadata-monolithic'
+}
+
+task downloadLibs(type: Sync) {
+    from configurations.libMetadata
+    into "$buildDir/libs"
+    rename 'tensorflow-lite-support-0.0.0-experimental-metadata-monolithic.jar', "tensorflow-lite-support-metadata.jar"
+}
+
+preBuild.dependsOn downloadLibs
+
+dependencies {
+    compileOnly 'org.checkerframework:checker-qual:2.5.8'
+    api 'org.tensorflow:tensorflow-lite:0.0.0-nightly'
+    api 'org.tensorflow:tensorflow-lite-support:0.0.0-nightly'
+    api files("$buildDir/libs/tensorflow-lite-support-metadata.jar")
+    implementation 'org.apache.commons:commons-compress:1.19'
+})");
+  return true;
+}
+
+bool GenerateAndroidManifestContent(CodeWriter* code_writer,
+                                    const ModelInfo& model_info) {
+  code_writer->Append(R"(<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="{{PACKAGE}}">
+</manifest>)");
+  return true;
+}
+
+bool GenerateDocContent(CodeWriter* code_writer, const ModelInfo& model_info) {
+  code_writer->Append("# {{MODEL_CLASS_NAME}} Usage");
+  // TODO(b/158651848) Generate imports for TFLS util types like TensorImage.
+  code_writer->AppendNoNewLine(R"(
+```
+import {{PACKAGE}}.{{MODEL_CLASS_NAME}};
+
+// 1. Initialize the Model
+{{MODEL_CLASS_NAME}} model = null;
+
+try {
+    model = {{MODEL_CLASS_NAME}}.newInstance(context);  // android.content.Context
+} catch (IOException e) {
+    e.printStackTrace();
+}
+
+if (model != null) {
+
+    // 2. Set the inputs)");
+  for (const auto& t : model_info.inputs) {
+    SetCodeWriterWithTensorInfo(code_writer, t);
+    if (t.content_type == "image") {
+      code_writer->Append(R"(
+    // Prepare tensor "{{NAME}}" from a Bitmap with ARGB_8888 format.
+    Bitmap bitmap = ...;
+    TensorImage {{NAME}} = TensorImage.fromBitmap(bitmap);
+    // Alternatively, load the input tensor "{{NAME}}" from pixel values.
+    // Check out TensorImage documentation to load other image data structures.
+    // int[] pixelValues = ...;
+    // int[] shape = ...;
+    // TensorImage {{NAME}} = new TensorImage();
+    // {{NAME}}.load(pixelValues, shape);)");
+    } else {
+      code_writer->Append(R"(
+    // Prepare input tensor "{{NAME}}" from an array.
+    // Check out TensorBuffer documentation to load other data structures.
+    TensorBuffer {{NAME}} = ...;
+    int[] values = ...;
+    int[] shape = ...;
+    {{NAME}}.load(values, shape);)");
+    }
+  }
+  code_writer->Append(R"(
+    // 3. Run the model
+    {{MODEL_CLASS_NAME}}.Outputs outputs = model.process({{INPUTS_LIST}});)");
+  code_writer->Append(R"(
+    // 4. Retrieve the results)");
+  for (const auto& t : model_info.outputs) {
+    SetCodeWriterWithTensorInfo(code_writer, t);
+    if (t.associated_axis_label_index >= 0) {
+      code_writer->SetTokenValue("WRAPPER_TYPE", "List<Category>");
+      code_writer->Append(
+          "    List<Category> {{NAME}} = "
+          "outputs.get{{NAME_U}}AsCategoryList();");
+    } else {
+      code_writer->Append(
+          "    {{WRAPPER_TYPE}} {{NAME}} = "
+          "outputs.get{{NAME_U}}As{{WRAPPER_TYPE}}();");
+    }
+  }
+  code_writer->Append(R"(}
+```)");
+  return true;
+}
+
+GenerationResult::File GenerateWrapperFile(const std::string& module_root,
+                                           const ModelInfo& model_info,
+                                           ErrorReporter* err) {
+  const auto java_path = JoinPath(module_root, "src/main/java");
+  const auto package_path =
+      JoinPath(java_path, ConvertPackageToPath(model_info.package_name));
+  const auto file_path =
+      JoinPath(package_path, model_info.model_class_name + JAVA_EXT);
+
+  CodeWriter code_writer(err);
+  code_writer.SetIndentString("  ");
+  SetCodeWriterWithModelInfo(&code_writer, model_info);
+
+  if (!GenerateWrapperFileContent(&code_writer, model_info, err)) {
+    err->Error("Generating Java wrapper content failed.");
+  }
+
+  const auto java_file = code_writer.ToString();
+  return GenerationResult::File{file_path, java_file};
+}
+
+GenerationResult::File GenerateBuildGradle(const std::string& module_root,
+                                           const ModelInfo& model_info,
+                                           ErrorReporter* err) {
+  const auto file_path = JoinPath(module_root, "build.gradle");
+  CodeWriter code_writer(err);
+  SetCodeWriterWithModelInfo(&code_writer, model_info);
+  if (!GenerateBuildGradleContent(&code_writer, model_info)) {
+    err->Error("Generating build.gradle failed.");
+  }
+  const auto content = code_writer.ToString();
+  return GenerationResult::File{file_path, content};
+}
+
+GenerationResult::File GenerateAndroidManifest(const std::string& module_root,
+                                               const ModelInfo& model_info,
+                                               ErrorReporter* err) {
+  const auto file_path = JoinPath(module_root, "src/main/AndroidManifest.xml");
+  CodeWriter code_writer(err);
+  SetCodeWriterWithModelInfo(&code_writer, model_info);
+  if (!GenerateAndroidManifestContent(&code_writer, model_info)) {
+    err->Error("Generating AndroidManifest.xml failed.");
+  }
+  return GenerationResult::File{file_path, code_writer.ToString()};
+}
+
+GenerationResult::File GenerateDoc(const std::string& module_root,
+                                   const ModelInfo& model_info,
+                                   ErrorReporter* err) {
+  std::string lower = model_info.model_class_name;
+  for (int i = 0; i < lower.length(); i++) {
+    lower[i] = std::tolower(lower[i]);
+  }
+  const auto file_path = JoinPath(module_root, lower + ".md");
+  CodeWriter code_writer(err);
+  SetCodeWriterWithModelInfo(&code_writer, model_info);
+  if (!GenerateDocContent(&code_writer, model_info)) {
+    err->Error("Generating doc failed.");
+  }
+  return GenerationResult::File{file_path, code_writer.ToString()};
+}
+
+}  // namespace
+
+AndroidJavaGenerator::AndroidJavaGenerator(const std::string& module_root)
+    : CodeGenerator(), module_root_(module_root) {}
+
+GenerationResult AndroidJavaGenerator::Generate(
+    const Model* model, const std::string& package_name,
+    const std::string& model_class_name, const std::string& model_asset_path) {
+  GenerationResult result;
+  if (model == nullptr) {
+    err_.Error(
+        "Cannot read model from the buffer. Codegen will generate nothing.");
+    return result;
+  }
+  const ModelMetadata* metadata = GetMetadataFromModel(model);
+  if (metadata == nullptr) {
+    err_.Error(
+        "Cannot find TFLite Metadata in the model. Codegen will generate "
+        "nothing.");
+    return result;
+  }
+  details_android_java::ModelInfo model_info = CreateModelInfo(
+      metadata, package_name, model_class_name, model_asset_path, &err_);
+  result.files.push_back(GenerateWrapperFile(module_root_, model_info, &err_));
+  result.files.push_back(GenerateBuildGradle(module_root_, model_info, &err_));
+  result.files.push_back(
+      GenerateAndroidManifest(module_root_, model_info, &err_));
+  result.files.push_back(GenerateDoc(module_root_, model_info, &err_));
+  return result;
+}
+
+GenerationResult AndroidJavaGenerator::Generate(
+    const char* model_storage, const std::string& package_name,
+    const std::string& model_class_name, const std::string& model_asset_path) {
+  const Model* model = GetModel(model_storage);
+  return Generate(model, package_name, model_class_name, model_asset_path);
+}
+
+std::string AndroidJavaGenerator::GetErrorMessage() {
+  return err_.GetMessage();
+}
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/codegen/android_java_generator.h b/tensorflow_lite_support/codegen/android_java_generator.h
new file mode 100644
index 0000000..634ccf6
--- /dev/null
+++ b/tensorflow_lite_support/codegen/android_java_generator.h
@@ -0,0 +1,116 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CODEGEN_ANDROID_JAVA_GENERATOR_H_
+#define TENSORFLOW_LITE_SUPPORT_CODEGEN_ANDROID_JAVA_GENERATOR_H_
+
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "tensorflow_lite_support/codegen/code_generator.h"
+#include "tensorflow_lite_support/codegen/utils.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+#include "tensorflow/lite/schema/schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+namespace details_android_java {
+
+/// The intermediate data structure for generating code from TensorMetadata.
+/// Should only be used as const reference when created.
+struct TensorInfo {
+  std::string name;
+  std::string upper_camel_name;
+  std::string content_type;
+  std::string wrapper_type;
+  std::string processor_type;
+  bool is_input;
+  /// Optional. Set to -1 if not applicable.
+  int normalization_unit;
+  /// Optional. Set to -1 if associated_axis_label is empty.
+  int associated_axis_label_index;
+  /// Optional. Set to -1 if associated_value_label is empty.
+  int associated_value_label_index;
+};
+
+/// The intermediate data structure for generating code from ModelMetadata.
+/// Should only be used as const reference when created.
+struct ModelInfo {
+  std::string package_name;
+  std::string model_asset_path;
+  std::string model_class_name;
+  std::string model_versioned_name;
+  std::vector<TensorInfo> inputs;
+  std::vector<TensorInfo> outputs;
+  // Extra helper fields. For models with inputs "a", "b" and outputs "x", "y":
+  std::string input_type_param_list;
+  // e.g. "TensorImage a, TensorBuffer b"
+  std::string inputs_list;
+  // e.g. "a, b"
+  std::string postprocessor_type_param_list;
+  // e.g. "ImageProcessor xPostprocessor, TensorProcessor yPostprocessor"
+  std::string postprocessors_list;
+  // e.g. "xPostprocessor, yPostprocessor"
+};
+
+}  // namespace details_android_java
+
+constexpr char JAVA_EXT[] = ".java";
+
+/// Generates Android supporting codes and modules (in Java) based on TFLite
+/// metadata.
+class AndroidJavaGenerator : public CodeGenerator {
+ public:
+  /// Creates an AndroidJavaGenerator.
+  /// Args:
+  /// - module_root: The root of destination Java module.
+  explicit AndroidJavaGenerator(const std::string& module_root);
+
+  /// Generates files. Returns the file paths and contents.
+  /// Args:
+  /// - model: The TFLite model with Metadata filled.
+  /// - package_name: The name of the Java package which generated classes
+  /// belong to.
+  /// - model_class_name: A readable name of the generated wrapper class, such
+  /// as "ImageClassifier", "MobileNetV2" or "MyModel".
+  /// - model_asset_path: The relevant path to the model file in the asset.
+  // TODO(b/141225157): Automatically generate model_class_name.
+  GenerationResult Generate(const Model* model, const std::string& package_name,
+                            const std::string& model_class_name,
+                            const std::string& model_asset_path);
+
+  /// Generates files and returns the file paths and contents.
+  /// It's mostly identical with the previous one, but the model here is
+  /// provided as binary flatbuffer content without parsing.
+  GenerationResult Generate(const char* model_storage,
+                            const std::string& package_name,
+                            const std::string& model_class_name,
+                            const std::string& model_asset_path);
+
+  std::string GetErrorMessage();
+
+ private:
+  const std::string module_root_;
+  ErrorReporter err_;
+};
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CODEGEN_ANDROID_JAVA_GENERATOR_H_
diff --git a/tensorflow_lite_support/codegen/code_generator.cc b/tensorflow_lite_support/codegen/code_generator.cc
new file mode 100644
index 0000000..1337708
--- /dev/null
+++ b/tensorflow_lite_support/codegen/code_generator.cc
@@ -0,0 +1,179 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/codegen/code_generator.h"
+
+#include <cctype>
+#include <unordered_map>
+#include <unordered_set>
+#include <utility>
+
+#include "tensorflow_lite_support/codegen/utils.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+namespace {
+
+void ResolveConflictedNamesByAddingIndex(std::vector<std::string>* names_ptr) {
+  auto& names = *names_ptr;
+  std::unordered_map<std::string, int> indexes;
+  std::unordered_map<std::string, int> first_appearance;
+  for (int i = 0; i < names.size(); i++) {
+    if (indexes.find(names[i]) == indexes.end()) {
+      indexes[names[i]] = 1;
+      first_appearance[names[i]] = i;
+    } else {
+      indexes[names[i]] += 1;
+      names[i].append(std::to_string(indexes[names[i]]));
+    }
+  }
+  for (const auto& it : first_appearance) {
+    const auto& name = it.first;
+    const auto i = it.second;
+    if (indexes[name] > 1) {
+      names[i].append("1");
+    }
+  }
+}
+
+}  // namespace
+
+CodeGenerator::CodeGenerator() {}
+
+bool CodeGenerator::VerifyMetadata(const ModelMetadata* metadata,
+                                   ErrorReporter* err) {
+  if (metadata == nullptr) {
+    err->Error("Loading nullptr is not allowed");
+    return false;
+  }
+  if (metadata->subgraph_metadata()->size() != 1) {
+    err->Error("Only exact 1 subgraph is supported");
+    return false;
+  }
+  return true;
+}
+
+std::pair<std::vector<std::string>, std::vector<std::string>>
+CodeGenerator::NameInputsAndOutputs(const TensorMetadataList* inputs,
+                                    const TensorMetadataList* outputs) {
+  std::vector<std::string> input_names;
+  std::vector<std::string> output_names;
+  if (inputs != nullptr) {
+    input_names.reserve(inputs->size());
+    for (const auto* tensor : *inputs) {
+      input_names.push_back(NameTensor(*tensor, "input"));
+    }
+  }
+  if (outputs != nullptr) {
+    output_names.reserve(outputs->size());
+    for (const auto* tensor : *outputs) {
+      output_names.push_back(NameTensor(*tensor, "output"));
+    }
+  }
+  // Solve conflict
+  ResolveConflictedInputAndOutputNames(&input_names, &output_names);
+  return std::make_pair(input_names, output_names);
+}
+
+std::string CodeGenerator::ConvertToValidName(const std::string& name) {
+  // lowercase all
+  std::string result = name;
+  for (int i = 0; i < result.size(); i++) {
+    result[i] = std::tolower(result[i]);
+  }
+  // replace all non-alpha or non-numeric with underscores, except underscore
+  // itself
+  for (int i = 0; i < result.size(); i++) {
+    if (result[i] != '_' && !std::isalnum(result[i])) {
+      result[i] = '_';
+    }
+  }
+  // remove leading underscores
+  int leading_underscores = 0;
+  while (leading_underscores < result.size() &&
+         result[leading_underscores] == '_') {
+    leading_underscores++;
+  }
+  result.erase(0, leading_underscores);
+  if (result.empty()) {
+    return "";
+  }
+  // first char should be alpha
+  if (std::isalpha(result[0])) {
+    return result;
+  }
+  return "tensor_" + result;
+}
+
+std::string CodeGenerator::NameTensor(const TensorMetadata& tensor,
+                                      const std::string& default_name) {
+  if (tensor.name() != nullptr && tensor.name()->size() > 0) {
+    // TODO(b/141225157) Validate tensor name. It should be in lower case.
+    auto suggested_name = ConvertToValidName(tensor.name()->str());
+    if (!suggested_name.empty()) {
+      return suggested_name;
+    }
+  }
+  auto* content = tensor.content();
+  if (content == nullptr || content->content_properties() == nullptr) {
+    return default_name;
+  }
+  switch (content->content_properties_type()) {
+    case ContentProperties_ImageProperties:
+      return "image";
+    case ContentProperties_FeatureProperties:
+      return "feature";
+    default:
+      return default_name;
+  }
+}
+
+void CodeGenerator::ResolveConflictedInputAndOutputNames(
+    std::vector<std::string>* inputs, std::vector<std::string>* outputs) {
+  std::unordered_set<std::string> io_conflict;
+  auto& input_names = *inputs;
+  auto& output_names = *outputs;
+  for (const auto& input : input_names) {
+    if (io_conflict.find(input) != io_conflict.end()) {
+      continue;
+    }
+    for (const auto& output : output_names) {
+      if (input == output) {
+        io_conflict.insert(input);
+        break;
+      }
+    }
+  }
+  for (int i = 0; i < input_names.size(); i++) {
+    if (io_conflict.find(input_names[i]) != io_conflict.end()) {
+      input_names[i] = "input_" + input_names[i];
+    }
+  }
+  for (int i = 0; i < output_names.size(); i++) {
+    if (io_conflict.find(output_names[i]) != io_conflict.end()) {
+      output_names[i] = "output_" + output_names[i];
+    }
+  }
+  // 2. Second, add index if input[i] == input[j]
+  ResolveConflictedNamesByAddingIndex(&input_names);
+  ResolveConflictedNamesByAddingIndex(&output_names);
+}
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/codegen/code_generator.h b/tensorflow_lite_support/codegen/code_generator.h
new file mode 100644
index 0000000..b557773
--- /dev/null
+++ b/tensorflow_lite_support/codegen/code_generator.h
@@ -0,0 +1,80 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_CODE_GENERATOR_H_
+#define TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_CODE_GENERATOR_H_
+
+#include <map>
+#include <memory>
+#include <sstream>
+#include <string>
+
+#include "tensorflow_lite_support/codegen/utils.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+struct GenerationResult {
+  struct File {
+    std::string path;
+    std::string content;
+  };
+  std::vector<File> files;
+};
+
+/// Defines language-independent codegen strategies, like class naming, .etc.
+/// Should not be used directly.
+class CodeGenerator {
+ public:
+  CodeGenerator();
+
+  using TensorMetadataList =
+      typename flatbuffers::Vector<flatbuffers::Offset<TensorMetadata>>;
+
+  virtual ~CodeGenerator() {}
+
+  // Strategies.
+  /// Names all the IO tensors. It's useful when they don't have names, or the
+  /// names have conflicts. We have to name every tensor for code generation.
+  // TODO(b/141225157): Add reserved keywords check.
+  static std::pair<std::vector<std::string>, std::vector<std::string>>
+  NameInputsAndOutputs(const TensorMetadataList* inputs,
+                       const TensorMetadataList* outputs);
+
+  /// Loads a metadata for code generation.
+  /// Returns false if the metadata is not good for generation.
+  static bool VerifyMetadata(const ModelMetadata* metadata, ErrorReporter* err);
+
+ protected:
+  /// Converts a name into a valid form. Rules:
+  /// - lower all letters.
+  /// - replace all non alphabet nor numeric characters with underscores.
+  /// - remove prefix underscores.
+  /// - add prefix if the leading character is a number.
+  /// Returns empty string if not possible.
+  static std::string ConvertToValidName(const std::string& name);
+  static std::string NameTensor(const TensorMetadata& tensor,
+                                const std::string& default_name);
+  static void ResolveConflictedInputAndOutputNames(
+      std::vector<std::string>* input, std::vector<std::string>* output);
+};
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_CODE_GENERATOR_H_
diff --git a/tensorflow_lite_support/codegen/code_generator_test.cc b/tensorflow_lite_support/codegen/code_generator_test.cc
new file mode 100644
index 0000000..5e9d64a
--- /dev/null
+++ b/tensorflow_lite_support/codegen/code_generator_test.cc
@@ -0,0 +1,126 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/codegen/code_generator.h"
+
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+
+namespace tflite {
+namespace support {
+namespace codegen {
+namespace {
+
+using ::testing::ElementsAreArray;
+
+class CodeGeneratorTest : public ::testing::Test {
+ public:
+  class TestingCodeGenerator : public CodeGenerator {
+   public:
+    explicit TestingCodeGenerator() : CodeGenerator() {}
+
+    // Make tested method public.
+    static std::string ConvertToValidName(const std::string& name) {
+      return CodeGenerator::ConvertToValidName(name);
+    }
+    static void ResolveConflictedInputAndOutputNames(
+        std::vector<std::string>* input, std::vector<std::string>* output) {
+      CodeGenerator::ResolveConflictedInputAndOutputNames(input, output);
+    }
+  };
+};
+
+TEST_F(CodeGeneratorTest, UpperCasesShouldLower) {
+  EXPECT_THAT(TestingCodeGenerator::ConvertToValidName("AlphaBetCOOL"),
+              "alphabetcool");
+}
+
+TEST_F(CodeGeneratorTest, NonAlphaNumShouldReplace) {
+  EXPECT_THAT(TestingCodeGenerator::ConvertToValidName("A+=B C\t"), "a__b_c_");
+}
+
+TEST_F(CodeGeneratorTest, NoLeadingUnderscore) {
+  EXPECT_THAT(TestingCodeGenerator::ConvertToValidName("+KAI Z"), "kai_z");
+}
+
+TEST_F(CodeGeneratorTest, NoLeadingNumbers) {
+  EXPECT_THAT(TestingCodeGenerator::ConvertToValidName("3000 Cool Tensors"),
+              "tensor_3000_cool_tensors");
+}
+
+TEST_F(CodeGeneratorTest, TestSimpleIONames) {
+  std::vector<std::string> inputs = {"image"};
+  std::vector<std::string> outputs = {"output"};
+  TestingCodeGenerator::ResolveConflictedInputAndOutputNames(&inputs, &outputs);
+  EXPECT_THAT(inputs, ElementsAreArray({"image"}));
+  EXPECT_THAT(outputs, ElementsAreArray({"output"}));
+}
+
+TEST_F(CodeGeneratorTest, TestIOConflict) {
+  std::vector<std::string> inputs = {"image"};
+  std::vector<std::string> outputs = {"image"};
+  TestingCodeGenerator::ResolveConflictedInputAndOutputNames(&inputs, &outputs);
+  EXPECT_THAT(inputs, ElementsAreArray({"input_image"}));
+  EXPECT_THAT(outputs, ElementsAreArray({"output_image"}));
+}
+
+TEST_F(CodeGeneratorTest, TestInternalConflict) {
+  std::vector<std::string> inputs = {"image", "image"};
+  std::vector<std::string> outputs = {"output"};
+  TestingCodeGenerator::ResolveConflictedInputAndOutputNames(&inputs, &outputs);
+  EXPECT_THAT(inputs, ElementsAreArray({"image1", "image2"}));
+  EXPECT_THAT(outputs, ElementsAreArray({"output"}));
+}
+
+TEST_F(CodeGeneratorTest, TestAllConflictNTo1) {
+  std::vector<std::string> inputs = {"image", "image"};
+  std::vector<std::string> outputs = {"image"};
+  TestingCodeGenerator::ResolveConflictedInputAndOutputNames(&inputs, &outputs);
+  EXPECT_THAT(inputs, ElementsAreArray({"input_image1", "input_image2"}));
+  EXPECT_THAT(outputs, ElementsAreArray({"output_image"}));
+}
+
+TEST_F(CodeGeneratorTest, TestAllConflict) {
+  std::vector<std::string> inputs = {"image", "audio", "image", "audio",
+                                     "audio"};
+  std::vector<std::string> outputs = {"image", "image", "audio", "feature",
+                                      "feature"};
+  TestingCodeGenerator::ResolveConflictedInputAndOutputNames(&inputs, &outputs);
+  EXPECT_THAT(inputs,
+              ElementsAreArray({"input_image1", "input_audio1", "input_image2",
+                                "input_audio2", "input_audio3"}));
+  EXPECT_THAT(outputs,
+              ElementsAreArray({"output_image1", "output_image2",
+                                "output_audio", "feature1", "feature2"}));
+}
+
+TEST_F(CodeGeneratorTest, TestAllConflictReversed) {
+  std::vector<std::string> inputs = {"image", "image", "audio", "feature",
+                                     "feature"};
+  std::vector<std::string> outputs = {"image", "audio", "image", "audio",
+                                      "audio"};
+  TestingCodeGenerator::ResolveConflictedInputAndOutputNames(&inputs, &outputs);
+  EXPECT_THAT(inputs,
+              ElementsAreArray({"input_image1", "input_image2", "input_audio",
+                                "feature1", "feature2"}));
+  EXPECT_THAT(outputs, ElementsAreArray({"output_image1", "output_audio1",
+                                         "output_image2", "output_audio2",
+                                         "output_audio3"}));
+}
+
+}  // namespace
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/codegen/metadata_helper.cc b/tensorflow_lite_support/codegen/metadata_helper.cc
new file mode 100644
index 0000000..00c9723
--- /dev/null
+++ b/tensorflow_lite_support/codegen/metadata_helper.cc
@@ -0,0 +1,100 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/codegen/metadata_helper.h"
+
+#include "tensorflow_lite_support/codegen/utils.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+constexpr char BUFFER_KEY[] = "TFLITE_METADATA";
+const ModelMetadata* GetMetadataFromModel(const Model* model) {
+  if (model == nullptr || model->metadata() == nullptr) {
+    return nullptr;
+  }
+  for (auto i = 0; i < model->metadata()->size(); i++) {
+    const auto* name = model->metadata()->Get(i)->name();
+    if (name != nullptr && name->str() == BUFFER_KEY) {
+      const auto buffer_index = model->metadata()->Get(i)->buffer();
+      if (model->buffers() == nullptr ||
+          model->buffers()->size() <= buffer_index) {
+        continue;
+      }
+      const auto* buffer_vec = model->buffers()->Get(buffer_index)->data();
+      if (buffer_vec == nullptr || buffer_vec->data() == nullptr) {
+        continue;
+      }
+      return GetModelMetadata(buffer_vec->data());
+    }
+  }
+  return nullptr;
+}
+
+int FindAssociatedFile(const TensorMetadata* metadata,
+                       const AssociatedFileType file_type,
+                       const std::string& tensor_identifier,
+                       ErrorReporter* err) {
+  int result = -1;
+  if (metadata->associated_files() == nullptr ||
+      metadata->associated_files()->size() == 0) {
+    return result;
+  }
+  for (int i = 0; i < metadata->associated_files()->size(); i++) {
+    const auto* file_metadata = metadata->associated_files()->Get(i);
+    if (file_metadata->type() == file_type) {
+      if (result >= 0) {
+        err->Warning(
+            "Multiple associated file of type %d found on tensor %s. Only the "
+            "first one will be used.",
+            file_type, tensor_identifier.c_str());
+        continue;
+      }
+      result = i;
+    }
+  }
+  return result;
+}
+
+int FindNormalizationUnit(const TensorMetadata* metadata,
+                          const std::string& tensor_identifier,
+                          ErrorReporter* err) {
+  int result = -1;
+  if (metadata->process_units() == nullptr ||
+      metadata->process_units()->size() == 0) {
+    return result;
+  }
+  for (int i = 0; i < metadata->process_units()->size(); i++) {
+    const auto* process_uint = metadata->process_units()->Get(i);
+    if (process_uint->options_type() ==
+        ProcessUnitOptions_NormalizationOptions) {
+      if (result >= 0) {
+        err->Warning(
+            "Multiple normalization unit found in tensor %s. Only the first "
+            "one will be effective.",
+            tensor_identifier.c_str());
+        continue;
+      }
+      result = i;
+    }
+  }
+  return result;
+}
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/codegen/metadata_helper.h b/tensorflow_lite_support/codegen/metadata_helper.h
new file mode 100644
index 0000000..8e3dc6a
--- /dev/null
+++ b/tensorflow_lite_support/codegen/metadata_helper.h
@@ -0,0 +1,51 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_METADATA_HELPER_H_
+#define TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_METADATA_HELPER_H_
+
+#include <string>
+
+#include "tensorflow_lite_support/codegen/utils.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+#include "tensorflow/lite/schema/schema_generated.h"
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+/// Parses a ModelMetadata out from a Model. The returned ModelMetadata's
+/// lifetime is scoped by the model.
+/// Returns nullptr if we cannot find any metadata.
+const ModelMetadata* GetMetadataFromModel(const Model* model);
+
+/// Finds an associated file from a TensorMetadata of certain type. If there're
+/// multiple files meet the criteria, only the first one is used. If there's no
+/// file meets the criteria, -1 will be returned.
+int FindAssociatedFile(const TensorMetadata* metadata,
+                       const AssociatedFileType file_type,
+                       const std::string& tensor_identifier,
+                       ErrorReporter* err);
+
+/// Find the first normalization unit. If none, return -1.
+int FindNormalizationUnit(const TensorMetadata* metadata,
+                          const std::string& tensor_identifier,
+                          ErrorReporter* err);
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_METADATA_HELPER_H_
diff --git a/tensorflow_lite_support/codegen/python/BUILD b/tensorflow_lite_support/codegen/python/BUILD
new file mode 100644
index 0000000..ee4dcbb
--- /dev/null
+++ b/tensorflow_lite_support/codegen/python/BUILD
@@ -0,0 +1,37 @@
+load("@org_tensorflow//tensorflow:tensorflow.bzl", "pybind_extension")
+
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+pybind_extension(
+    name = "_pywrap_codegen",
+    srcs = [
+        "codegen_lib.cc",
+    ],
+    features = ["-use_header_modules"],
+    module_name = "_pywrap_codegen",
+    deps = [
+        "//tensorflow_lite_support/codegen:android_java_generator",
+        "//tensorflow_lite_support/codegen:code_generator",
+        "@local_config_python//:python_headers",
+        "@pybind11",
+    ],
+)
+
+py_binary(
+    name = "codegen",
+    srcs = [
+        "codegen.py",
+    ],
+    python_version = "PY3",
+    deps = [
+        ":_pywrap_codegen",
+        "@absl_py//absl:app",
+        "@absl_py//absl/flags",
+        "@absl_py//absl/logging",
+    ],
+)
diff --git a/tensorflow_lite_support/codegen/python/codegen.py b/tensorflow_lite_support/codegen/python/codegen.py
new file mode 100644
index 0000000..7309a69
--- /dev/null
+++ b/tensorflow_lite_support/codegen/python/codegen.py
@@ -0,0 +1,104 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""Generates Android Java sources from a TFLite model with metadata."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import shutil
+import sys
+from absl import app
+from absl import flags
+from absl import logging
+
+from tensorflow_lite_support.codegen.python import _pywrap_codegen
+
+FLAGS = flags.FLAGS
+
+flags.DEFINE_string('model', None, 'Path to model (.tflite) flatbuffer file.')
+flags.DEFINE_string('destination', None, 'Path of destination of generation.')
+flags.DEFINE_string('package_name', 'org.tensorflow.lite.support',
+                    'Name of generated java package to put the wrapper class.')
+flags.DEFINE_string(
+    'model_class_name', 'MyModel',
+    'Name of generated wrapper class (should not contain package name).')
+flags.DEFINE_string(
+    'model_asset_path', '',
+    '(Optional) Path to the model in generated assets/ dir. If not set, '
+    'generator will use base name of input model.'
+)
+
+
+def get_model_buffer(path):
+  if not os.path.isfile(path):
+    logging.error('Cannot find model at path %s.', path)
+  with open(path, 'rb') as f:
+    buf = f.read()
+    return buf
+
+
+def prepare_directory_for_file(file_path):
+  target_dir = os.path.dirname(file_path)
+  if not os.path.exists(target_dir):
+    os.makedirs(target_dir)
+    return
+  if not os.path.isdir(target_dir):
+    logging.error('Cannot write to %s', target_dir)
+
+
+def run_main(argv):
+  """Main function of the codegen."""
+
+  if len(argv) > 1:
+    logging.error('None flag arguments found: [%s]', ', '.join(argv[1:]))
+
+  codegen = _pywrap_codegen.AndroidJavaGenerator(FLAGS.destination)
+  model_buffer = get_model_buffer(FLAGS.model)
+  model_asset_path = FLAGS.model_asset_path
+  if not model_asset_path:
+    model_asset_path = os.path.basename(FLAGS.model)
+  result = codegen.generate(model_buffer, FLAGS.package_name,
+                            FLAGS.model_class_name, model_asset_path)
+  error_message = codegen.get_error_message().strip()
+  if error_message:
+    logging.error(error_message)
+  if not result.files:
+    logging.error('Generation failed!')
+    return
+
+  for each in result.files:
+    prepare_directory_for_file(each.path)
+    with open(each.path, 'w') as f:
+      f.write(each.content)
+
+  logging.info('Generation succeeded!')
+  model_asset_path = os.path.join(FLAGS.destination, 'src/main/assets',
+                                  model_asset_path)
+  prepare_directory_for_file(model_asset_path)
+  shutil.copy(FLAGS.model, model_asset_path)
+  logging.info('Model copied into assets!')
+
+
+# Simple wrapper to make the code pip-friendly
+def main():
+  flags.mark_flag_as_required('model')
+  flags.mark_flag_as_required('destination')
+  app.run(main=run_main, argv=sys.argv)
+
+
+if __name__ == '__main__':
+  app.run(main)
diff --git a/tensorflow_lite_support/codegen/python/codegen_lib.cc b/tensorflow_lite_support/codegen/python/codegen_lib.cc
new file mode 100644
index 0000000..6b2cd5e
--- /dev/null
+++ b/tensorflow_lite_support/codegen/python/codegen_lib.cc
@@ -0,0 +1,49 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "pybind11/detail/common.h"
+#include "pybind11/pybind11.h"
+#include "pybind11/pytypes.h"
+#include "pybind11/stl.h"
+#include "tensorflow_lite_support/codegen/android_java_generator.h"
+#include "tensorflow_lite_support/codegen/code_generator.h"
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+template <typename... Args>
+using overload_cast_ = pybind11::detail::overload_cast_impl<Args...>;
+
+PYBIND11_MODULE(_pywrap_codegen, m) {
+  pybind11::class_<AndroidJavaGenerator>(m, "AndroidJavaGenerator")
+      .def(pybind11::init<const std::string &>())
+      .def("generate",
+           overload_cast_<const char *, const std::string &,
+                          const std::string &, const std::string &>()(
+               &AndroidJavaGenerator::Generate))
+      .def("get_error_message", &AndroidJavaGenerator::GetErrorMessage);
+  pybind11::class_<GenerationResult>(m, "GenerationResult")
+      .def(pybind11::init<>())
+      .def_readwrite("files", &GenerationResult::files);
+  pybind11::class_<GenerationResult::File>(m, "GenerationResultFile")
+      .def(pybind11::init<>())
+      .def_readwrite("path", &GenerationResult::File::path)
+      .def_readwrite("content", &GenerationResult::File::content);
+}
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/codegen/utils.cc b/tensorflow_lite_support/codegen/utils.cc
new file mode 100644
index 0000000..c75fc5f
--- /dev/null
+++ b/tensorflow_lite_support/codegen/utils.cc
@@ -0,0 +1,194 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/codegen/utils.h"
+
+#include <cstdarg>
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+int ErrorReporter::Warning(const char* format, ...) {
+  va_list args;
+  va_start(args, format);
+  return Report("[WARN] ", format, args);
+}
+
+int ErrorReporter::Error(const char* format, ...) {
+  va_list args;
+  va_start(args, format);
+  return Report("[ERROR] ", format, args);
+}
+
+int ErrorReporter::Report(const char* prefix, const char* format,
+                          va_list args) {
+  char buf[1024];
+  int formatted = vsnprintf(buf, sizeof(buf), format, args);
+  buffer_ << prefix << buf << std::endl;
+  return formatted;
+}
+
+std::string ErrorReporter::GetMessage() {
+  std::string value = buffer_.str();
+  buffer_.str("");
+  return value;
+}
+
+CodeWriter::CodeWriter(ErrorReporter* err) : indent_(0), err_(err) {}
+
+void CodeWriter::SetTokenValue(const std::string& token,
+                               const std::string& value) {
+  value_map_[token] = value;
+}
+
+const std::string CodeWriter::GetTokenValue(const std::string& token) const {
+  auto iter = value_map_.find(token);
+  if (iter == value_map_.end()) {
+    // Typically only Code Generator's call this function (or `Append`). It's
+    // their duty to make sure the token is valid, and requesting for an invalid
+    // token implicits flaws in the code generation logic.
+    err_->Error("Internal: Cannot find value with token '%s'", token.c_str());
+    return "";
+  }
+  return iter->second;
+}
+
+void CodeWriter::SetIndentString(const std::string& indent_str) {
+  indent_str_ = indent_str;
+}
+
+void CodeWriter::Indent() { indent_++; }
+
+void CodeWriter::Outdent() { indent_--; }
+
+std::string CodeWriter::GenerateIndent() const {
+  std::string res;
+  res.reserve(indent_str_.size() * indent_);
+  for (int i = 0; i < indent_; i++) {
+    res.append(indent_str_);
+  }
+  return res;
+}
+
+void CodeWriter::Append(const std::string& text) { AppendInternal(text, true); }
+
+void CodeWriter::AppendNoNewLine(const std::string& text) {
+  AppendInternal(text, false);
+}
+
+void CodeWriter::AppendInternal(const std::string& text, bool newline) {
+  // Prefix indent
+  if ((buffer_.empty()             // nothing in the buffer
+       || buffer_.back() == '\n')  // is on new line
+      // is writing on current line
+      && (!text.empty() && text[0] != '\n' && text[0] != '\r')) {
+    buffer_.append(GenerateIndent());
+  }
+  // State machine variables
+  bool in_token = false;
+  int i = 0;
+  // Rough memory reserve
+  buffer_.reserve(buffer_.size() + text.size());
+  std::string token_buffer;
+  // A simple LL1 analysis
+  while (i < text.size()) {
+    char cur = text[i];
+    char cur_next = i == text.size() - 1 ? '\0' : text[i + 1];  // Set guardian
+    if (!in_token) {
+      if (cur == '{' && cur_next == '{') {  // Enter token
+        in_token = true;
+        i += 2;
+      } else if (cur == '\n') {  // We need to apply global indent here
+        buffer_.push_back(cur);
+        if (cur_next != '\0' && cur_next != '\n' && cur_next != '\r') {
+          buffer_.append(GenerateIndent());
+        }
+        i += 1;
+      } else {
+        buffer_.push_back(cur);
+        i += 1;
+      }
+    } else {
+      if (cur == '}' && cur_next == '}') {  // Close token
+        in_token = false;
+        const auto value = GetTokenValue(token_buffer);
+        buffer_.append(value);
+        token_buffer.clear();
+        i += 2;
+      } else {
+        token_buffer.push_back(cur);
+        i += 1;
+      }
+    }
+  }
+  if (!token_buffer.empty()) {
+    // Typically only Code Generator's call this function. It's
+    // their duty to make sure the code (or template) has valid syntax, and
+    // unclosed "{{...}}" implicits severe error in the template.
+    err_->Error("Internal: Invalid template: {{token}} is not closed.");
+  }
+  if (newline) {
+    buffer_.push_back('\n');
+  }
+}
+
+void CodeWriter::NewLine() { Append(""); }
+
+void CodeWriter::Backspace(int n) {
+  buffer_.resize(buffer_.size() > n ? buffer_.size() - n : 0);
+}
+
+std::string CodeWriter::ToString() const { return buffer_; }
+
+bool CodeWriter::IsStreamEmpty() const { return buffer_.empty(); }
+
+void CodeWriter::Clear() {
+  buffer_.clear();
+  value_map_.clear();
+  indent_ = 0;
+}
+
+std::string SnakeCaseToCamelCase(const std::string& s) {
+  std::string t;
+  t.reserve(s.length());
+  size_t i = 0;
+  // Note: Use simple string += for simplicity.
+  bool cap = false;
+  while (i < s.size()) {
+    const char c = s[i++];
+    if (c == '_') {
+      cap = true;
+    } else if (cap) {
+      t += toupper(c);
+      cap = false;
+    } else {
+      t += c;
+    }
+  }
+  return t;
+}
+
+std::string JoinPath(const std::string& a, const std::string& b) {
+  if (a.empty()) return b;
+  std::string a_fixed = a;
+  if (!a_fixed.empty() && a_fixed.back() == '/') a_fixed.pop_back();
+  std::string b_fixed = b;
+  if (!b_fixed.empty() && b_fixed.front() == '/') b_fixed.erase(0, 1);
+  return a_fixed + "/" + b_fixed;
+}
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/codegen/utils.h b/tensorflow_lite_support/codegen/utils.h
new file mode 100644
index 0000000..98768b6
--- /dev/null
+++ b/tensorflow_lite_support/codegen/utils.h
@@ -0,0 +1,127 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_UTILS_H_
+#define TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_UTILS_H_
+
+#include <map>
+#include <sstream>
+#include <string>
+
+namespace tflite {
+namespace support {
+namespace codegen {
+
+/// Collects runtime error logs which could be showed later.
+// TODO(b/150538286): Consider a better mechanism to simplify callsite code.
+class ErrorReporter {
+ public:
+  int Warning(const char* format, ...);
+  int Error(const char* format, ...);
+  std::string GetMessage();
+
+ private:
+  int Report(const char* prefix, const char* format, va_list args);
+  std::stringstream buffer_;
+};
+
+/// Implements basic code generating with text templates.
+///
+/// It could accept code templates and concatenate them into complete codes. A
+/// template could contain named values.
+///
+/// Example code:
+///   CodeWriter code;
+///   code.SetValue("NAME", "Foo");
+///   code.Append("void {{NAME}}() { printf("%s", "{{NAME}}"); }");
+///   code.SetValue("NAME", "Bar");
+///   code.Append("void {{NAME}}() { printf("%s", "{{NAME}}"); }");
+///
+/// Output:
+///  void Foo() { printf("%s", "Foo"); }
+///  void Bar() { printf("%s", "Bar"); }
+class CodeWriter {
+ public:
+  explicit CodeWriter(ErrorReporter* err);
+  /// Sets value to a token. When generating code with template, a string in a
+  /// pair of {{ and }} will be regarded as a token and replaced with the
+  /// corresponding value in code generation.
+  /// It rewrites if the token already has a value.
+  void SetTokenValue(const std::string& token, const std::string& value);
+
+  /// Gets the current value set on the given token.
+  const std::string GetTokenValue(const std::string& token) const;
+
+  /// Sets the unit indent string. For example, in Java it should be "  ".
+  void SetIndentString(const std::string& indent);
+
+  /// Increases the indent by a unit (the string set in SetIndentString).
+  void Indent();
+
+  /// Decreases the indent by a unit (the string set in SetIndentString).
+  void Outdent();
+
+  /// Generates the indentation string.
+  std::string GenerateIndent() const;
+
+  /// Appends a piece of template codes to the stream. Every named value will be
+  /// replaced via the real value. A new line will always be appended at the
+  /// end.
+  void Append(const std::string& text);
+
+  /// Appends a piece of template codes to the stream. Same with `Append`, but a
+  /// new line will not be appended at the end.
+  void AppendNoNewLine(const std::string& text);
+
+  /// Appends a new line to the stream.
+  void NewLine();
+
+  /// Deletes the last N charaters in the stream. If the stream has less than N
+  /// characters, deletes all.
+  void Backspace(int n);
+
+  std::string ToString() const;
+
+  /// Checks if the internal string stream is empty. Note: This method has
+  // overhead.
+  bool IsStreamEmpty() const;
+
+  /// Clears all the internal string stream and value map.
+  void Clear();
+
+ private:
+  void AppendInternal(const std::string& text, bool newline);
+
+  std::string indent_str_;
+  int indent_;
+
+  std::map<std::string, std::string> value_map_;
+  std::string buffer_;
+
+  ErrorReporter* err_;
+};
+
+/// Converts foo_bar_name to fooBarName. It's callers duty to make sure given
+/// string "s" is already in snake case; or unexpected behavior may occur.
+std::string SnakeCaseToCamelCase(const std::string& s);
+
+/// Joins 2 parts of file path into one, connected by unix path seperator '/'.
+/// It's callers duty to ensure the two parts are valid.
+std::string JoinPath(const std::string& a, const std::string& b);
+
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
+
+#endif  // TENSORFLOW_TENSORFLOW_LITE_SUPPORT_CODEGEN_UTILS_H_
diff --git a/tensorflow_lite_support/codegen/utils_test.cc b/tensorflow_lite_support/codegen/utils_test.cc
new file mode 100644
index 0000000..3111f37
--- /dev/null
+++ b/tensorflow_lite_support/codegen/utils_test.cc
@@ -0,0 +1,97 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/codegen/utils.h"
+
+#include <gtest/gtest.h>
+
+namespace tflite {
+namespace support {
+namespace codegen {
+namespace {
+
+TEST(ErrorReporterTest, TestReportError) {
+  ErrorReporter err;
+  err.Error("some text");
+  EXPECT_EQ(err.GetMessage(), "[ERROR] some text\n");
+  EXPECT_EQ(err.GetMessage(), "");
+}
+
+TEST(CodeGeneratorTest, TestExample) {
+  ErrorReporter err;
+  CodeWriter writer(&err);
+  writer.SetTokenValue("NAME", "Foo");
+  const std::string text = R"(void {{NAME}}() { printf("%s", "{{NAME}}"); })";
+  writer.Append(text);
+  writer.SetTokenValue("NAME", "Bar");
+  writer.Append(text);
+  EXPECT_EQ(
+      "void Foo() { printf(\"%s\", \"Foo\"); }\n"
+      "void Bar() { printf(\"%s\", \"Bar\"); }\n",
+      writer.ToString());
+}
+
+TEST(CodeGeneratorTest, TestInexistentToken) {
+  ErrorReporter err;
+  CodeWriter writer(&err);
+  writer.SetTokenValue("NAME", "Foo");
+  const std::string text = R"(void {{name}}() {})";
+  writer.Append(text);
+  EXPECT_EQ(err.GetMessage(),
+            "[ERROR] Internal: Cannot find value with token 'name'\n");
+}
+
+TEST(CodeGeneratorTest, TestUnclosedToken) {
+  ErrorReporter err;
+  CodeWriter writer(&err);
+  writer.SetTokenValue("NAME", "Foo");
+  const std::string text = R"(void {{NAME}() {})";
+  writer.Append(text);
+  EXPECT_EQ(err.GetMessage(),
+            "[ERROR] Internal: Invalid template: {{token}} is not closed.\n");
+}
+
+TEST(CodeGeneratorTest, TestIndentControl) {
+  ErrorReporter err;
+  CodeWriter writer(&err);
+  writer.SetIndentString("  ");
+  writer.Indent();
+  writer.AppendNoNewLine("abcde");  // Will indent
+  EXPECT_EQ("  abcde", writer.ToString());
+  writer.Clear();
+  writer.Indent();
+  writer.AppendNoNewLine("abc\n\nde");
+  // The blank line will not indent
+  EXPECT_EQ("  abc\n\n  de", writer.ToString());
+  writer.Clear();
+  writer.Indent();
+  writer.Append("abc");
+  writer.Outdent();
+  writer.AppendNoNewLine("def");
+  EXPECT_EQ("  abc\ndef", writer.ToString());
+}
+
+TEST(CaseConversionTest, TestSnakeToCamel) {
+  EXPECT_EQ("imACamel", SnakeCaseToCamelCase("im_a_camel"));
+  EXPECT_EQ("imACamel", SnakeCaseToCamelCase("im_a_camel_"));
+  EXPECT_EQ("ImACamel", SnakeCaseToCamelCase("_im_a_camel"));
+  EXPECT_EQ("", SnakeCaseToCamelCase("_"));
+  EXPECT_EQ("camel", SnakeCaseToCamelCase("camel"));
+}
+
+}  // namespace
+}  // namespace codegen
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/BUILD b/tensorflow_lite_support/custom_ops/BUILD
new file mode 100644
index 0000000..35734cf
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/BUILD
@@ -0,0 +1,43 @@
+load("@org_tensorflow//tensorflow/lite/delegates/flex:build_def.bzl", "tflite_flex_cc_library")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+# This will generate the tf_text_flex_delegate cc_library, which is a custom
+# flex delegate that only contains ops in listed models.
+tflite_flex_cc_library(
+    name = "tf_text_flex_delegate",
+    additional_deps = ["@org_tensorflow_text//tensorflow_text:ops_lib"],
+    models = [
+        # TODO(b/160817619) Replace with a more complex model.
+        "testdata/sentencepiece_tokenizer_flex_op.tflite",
+    ],
+)
+
+# bazel test --config=monolithic tensorflow_lite_support/custom_ops:tflite_inference_test
+cc_test(
+    name = "tflite_inference_test",
+    srcs = ["tflite_inference_main.cc"],
+    args = ["--model=tensorflow_lite_support/custom_ops/testdata/sentencepiece_tokenizer_flex_op.tflite"],
+    data = ["//tensorflow_lite_support/custom_ops:testdata/sentencepiece_tokenizer_flex_op.tflite"],
+    deps = [
+        ":tf_text_flex_delegate",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite:string_util",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+        "@org_tensorflow//tensorflow/lite/tools:command_line_flags",
+    ] + select({
+        "@org_tensorflow//tensorflow:android": [
+            "@org_tensorflow//tensorflow/core:portable_tensorflow_lib_lite",
+        ],
+        "@org_tensorflow//tensorflow:ios": [
+            "@org_tensorflow//tensorflow/core:portable_tensorflow_lib_lite",
+        ],
+        "//conditions:default": [
+            "@org_tensorflow//tensorflow/core:lib",
+        ],
+    }),
+)
diff --git a/tensorflow_lite_support/custom_ops/kernel/BUILD b/tensorflow_lite_support/custom_ops/kernel/BUILD
new file mode 100644
index 0000000..b9b11de
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/BUILD
@@ -0,0 +1,146 @@
+# Placeholder for internal Python strict test compatibility macro.
+load("@org_tensorflow//tensorflow:tensorflow.bzl", "pybind_extension")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "whitespace_tokenizer",
+    srcs = ["whitespace_tokenizer.cc"],
+    hdrs = ["whitespace_tokenizer.h"],
+    deps = [
+        "@org_tensorflow//tensorflow/lite:context",
+        "@org_tensorflow//tensorflow/lite:string_util",
+        "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+        "@utf_archive//:utf",
+    ],
+)
+
+cc_library(
+    name = "whitespace_tokenizer_op_resolver",
+    srcs = ["whitespace_tokenizer_op_resolver.cc"],
+    hdrs = ["whitespace_tokenizer_op_resolver.h"],
+    visibility = ["//visibility:public"],
+    deps = [
+        ":whitespace_tokenizer",
+        "@org_tensorflow//tensorflow/lite:framework",
+    ],
+)
+
+pybind_extension(
+    name = "_pywrap_whitespace_tokenizer_op_resolver",
+    srcs = ["whitespace_tokenizer_op_resolver_wrapper.cc"],
+    hdrs = ["whitespace_tokenizer_op_resolver.h"],
+    additional_exported_symbols = ["AddWhitespaceTokenizerCustomOp"],
+    module_name = "_pywrap_whitespace_tokenizer_op_resolver",
+    visibility = ["//visibility:public"],
+    deps = [
+        ":whitespace_tokenizer_op_resolver",
+        "@local_config_python//:python_headers",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@pybind11",
+    ],
+)
+
+cc_test(
+    name = "whitespace_tokenizer_test",
+    srcs = ["whitespace_tokenizer_test.cc"],
+    deps = [
+        ":whitespace_tokenizer",
+        "@com_google_googletest//:gtest_main",
+        "@org_tensorflow//tensorflow/lite:string_util",
+        "@org_tensorflow//tensorflow/lite/kernels:test_util",
+        "@org_tensorflow//tensorflow/lite/schema:schema_fbs",
+    ],
+)
+
+py_test(
+    name = "whitespace_tokenizer_py_test",
+    srcs = ["whitespace_tokenizer_test.py"],
+    data = [
+        "testdata/whitespace_tokenizer_flex_delegate.tflite",
+        "testdata/whitespace_tokenizer_to_ragged_1d_input.tflite",
+        "testdata/whitespace_tokenizer_to_ragged_2d_input.tflite",
+        "testdata/whitespace_tokenizer_to_tensor.tflite",
+    ],
+    main = "whitespace_tokenizer_test.py",
+    python_version = "PY3",
+    srcs_version = "PY3",
+    deps = [
+        ":_pywrap_whitespace_tokenizer_op_resolver",
+        # numpy dep,
+        # tensorflow dep,
+        # tensorflow_text dep,
+        "@absl_py//absl/logging",
+        "@absl_py//absl/testing:parameterized",
+    ],
+)
+
+cc_library(
+    name = "ngrams",
+    srcs = ["ngrams.cc"],
+    hdrs = ["ngrams.h"],
+    deps = [
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:context",
+        "@org_tensorflow//tensorflow/lite:string_util",
+        "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+    ],
+)
+
+cc_library(
+    name = "ngrams_op_resolver",
+    srcs = ["ngrams_op_resolver.cc"],
+    hdrs = ["ngrams_op_resolver.h"],
+    visibility = ["//visibility:public"],
+    deps = [
+        ":ngrams",
+        "@org_tensorflow//tensorflow/lite:framework",
+    ],
+)
+
+pybind_extension(
+    name = "_pywrap_ngrams_op_resolver",
+    srcs = ["ngrams_op_resolver_wrapper.cc"],
+    hdrs = ["ngrams_op_resolver.h"],
+    additional_exported_symbols = ["AddNgramsCustomOp"],
+    module_name = "_pywrap_ngrams_op_resolver",
+    visibility = ["//visibility:public"],
+    deps = [
+        ":ngrams_op_resolver",
+        "@local_config_python//:python_headers",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@pybind11",
+    ],
+)
+
+cc_test(
+    name = "ngrams_test",
+    srcs = ["ngrams_test.cc"],
+    deps = [
+        ":ngrams",
+        "@com_google_googletest//:gtest_main",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:string_util",
+        "@org_tensorflow//tensorflow/lite/kernels:test_util",
+        "@org_tensorflow//tensorflow/lite/schema:schema_fbs",
+    ],
+)
+
+py_test(
+    name = "ngrams_py_test",
+    srcs = ["ngrams_test.py"],
+    main = "ngrams_test.py",
+    python_version = "PY3",
+    srcs_version = "PY3",
+    deps = [
+        ":_pywrap_ngrams_op_resolver",
+        # tensorflow dep,
+        # tensorflow_text dep,
+        "//tensorflow_lite_support/custom_ops/python:tflite_text_api",
+        "@absl_py//absl/logging",
+        "@absl_py//absl/testing:parameterized",
+    ],
+)
diff --git a/tensorflow_lite_support/custom_ops/kernel/ngrams.cc b/tensorflow_lite_support/custom_ops/kernel/ngrams.cc
new file mode 100644
index 0000000..3831c63
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ngrams.cc
@@ -0,0 +1,208 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/ngrams.h"
+
+#include "flatbuffers/flexbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/context.h"
+#include "tensorflow/lite/kernels/kernel_util.h"
+#include "tensorflow/lite/string_util.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace ngrams {
+
+// This TFLite op implements the text.ngrams when reduction_type = STRING_JOIN.
+//
+// Input:
+// * data: A string tensor, or a ragged string tensor (a 1D string value tensor
+//     and one or more 1D int64 row_split tensors).
+//
+// Attributes:
+// * width:             scalar integer
+//     The width of the ngram window.
+// * axis:              scalar integer
+//     The axis to create ngrams along.  For STRING_JOIN, this must be -1.
+// * reduction_type:    scalar string
+//     A string corresponding to the name of an enum value of text.Reduction
+//     Currently, only STRING_JOIN is supported.
+// * string_separator:  scalar string
+//     The separator string used to join tokens together.
+//
+// Output:
+// * output: A string tensor that matches the rank of 'data'.  Will be a ragged
+//     tensor if 'data' is a ragged tensor.
+
+// Both the input and output tensors use the same indices.
+constexpr int kValues = 0;
+constexpr int kRowSplitsStart = 1;
+
+// Reduction types.
+constexpr char kStringJoin[] = "STRING_JOIN";
+
+struct NgramsAttributes {
+  int width;
+  int axis;
+  std::string reduction_type;
+  std::string string_separator;
+
+  explicit NgramsAttributes(const flexbuffers::Map& m)
+      : width(m["width"].AsInt32()),
+        axis(m["axis"].AsInt32()),
+        reduction_type(m["reduction_type"].ToString()),
+        string_separator(m["string_separator"].ToString()) {}
+};
+
+inline bool OutputIsTensor(TfLiteNode* node) { return NumOutputs(node) == 1; }
+inline int NumRowSplits(TfLiteNode* node) {
+  return NumInputs(node) - kRowSplitsStart;
+}
+
+void* Init(TfLiteContext* context, const char* buffer, size_t length) {
+  const uint8_t* buffer_t = reinterpret_cast<const uint8_t*>(buffer);
+  return new NgramsAttributes(flexbuffers::GetRoot(buffer_t, length).AsMap());
+}
+
+void Free(TfLiteContext* context, void* buffer) {
+  delete reinterpret_cast<NgramsAttributes*>(buffer);
+}
+
+TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
+  const auto& attributes =
+      *reinterpret_cast<NgramsAttributes*>(node->user_data);
+
+  TF_LITE_ENSURE(context, attributes.reduction_type == kStringJoin);
+  TF_LITE_ENSURE(context, attributes.axis == -1);
+
+  TfLiteTensor* output_values = GetOutput(context, node, kValues);
+  if (OutputIsTensor(node)) {
+    const TfLiteTensor* input_values = GetInput(context, node, kValues);
+    int values_num_dims = NumDimensions(input_values);
+    TfLiteIntArray* output_values_shape = TfLiteIntArrayCreate(values_num_dims);
+    for (int i = 0; i < values_num_dims; ++i) {
+      output_values_shape->data[i] = SizeOfDimension(input_values, i);
+    }
+    output_values_shape->data[values_num_dims - 1] =
+        std::max(0, SizeOfDimension(input_values, values_num_dims - 1) -
+                        attributes.width + 1);
+    TF_LITE_ENSURE_OK(context, context->ResizeTensor(context, output_values,
+                                                     output_values_shape));
+    return kTfLiteOk;
+  }
+
+  SetTensorToDynamic(output_values);
+  // The row_splits tensors maintain their shape, because only the
+  // innermost dimension will change.
+  for (int i = kRowSplitsStart; i < NumOutputs(node); ++i) {
+    const TfLiteTensor* input_row_splits = GetInput(context, node, i);
+    TfLiteTensor* output_row_splits = GetOutput(context, node, i);
+    TF_LITE_ENSURE_EQ(context, NumDimensions(input_row_splits), 1);
+    TfLiteIntArray* output_row_splits_shape = TfLiteIntArrayCreate(1);
+    output_row_splits_shape->data[0] = SizeOfDimension(input_row_splits, 0);
+    TF_LITE_ENSURE_OK(context, context->ResizeTensor(context, output_row_splits,
+                                                     output_row_splits_shape));
+  }
+  return kTfLiteOk;
+}
+
+TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
+  const auto& attributes =
+      *reinterpret_cast<NgramsAttributes*>(node->user_data);
+
+  // Storage for the dummy input and output row_splits used in the tensor case.
+  std::vector<int64_t> tensor_input_row_splits;
+  std::vector<int64_t> tensor_output_row_splits;
+
+  const int64_t* input_row_splits;
+  int64_t* output_row_splits;
+  int n_row_splits = 0;
+
+  const TfLiteTensor* input_values = GetInput(context, node, kValues);
+
+  if (OutputIsTensor(node)) {
+    // Generate mock input and output innermost row_splits.
+    int64_t total_tokens = NumElements(input_values);
+    int64_t tokens_per_element =
+        SizeOfDimension(input_values, NumDimensions(input_values) - 1);
+    tensor_input_row_splits.reserve(total_tokens / tokens_per_element + 1);
+    tensor_output_row_splits.resize(total_tokens / tokens_per_element + 1);
+    for (int64_t i = 0; i <= total_tokens; i += tokens_per_element) {
+      tensor_input_row_splits.push_back(i);
+    }
+    input_row_splits = tensor_input_row_splits.data();
+    output_row_splits = tensor_output_row_splits.data();
+    n_row_splits = tensor_input_row_splits.size();
+  } else {
+    int index = 0;
+    while (index < NumRowSplits(node) - 1) {
+      const TfLiteTensor* input_tensor_row_splits =
+          GetInput(context, node, kRowSplitsStart + index);
+      TfLiteTensor* output_tensor_row_splits =
+          GetOutput(context, node, kRowSplitsStart + index);
+      memcpy(output_tensor_row_splits->data.raw,
+             input_tensor_row_splits->data.raw, input_tensor_row_splits->bytes);
+      ++index;
+    }
+
+    const TfLiteTensor* input_tensor_row_splits =
+        GetInput(context, node, kRowSplitsStart + index);
+    TfLiteTensor* output_tensor_row_splits =
+        GetOutput(context, node, kRowSplitsStart + index);
+    input_row_splits = input_tensor_row_splits->data.i64;
+    output_row_splits = output_tensor_row_splits->data.i64;
+    n_row_splits = SizeOfDimension(input_tensor_row_splits, 0);
+  }
+
+  DynamicBuffer buffer;
+  StringRef separator;
+  separator.str = attributes.string_separator.c_str();
+  separator.len = attributes.string_separator.length();
+  int buffer_index = 0;
+  for (int i = 0; i < n_row_splits - 1; ++i) {
+    output_row_splits[i] = buffer_index;
+    std::vector<StringRef> tokens;
+    for (int j = input_row_splits[i]; j < input_row_splits[i + 1]; ++j) {
+      tokens.emplace_back(GetString(input_values, j));
+      if (tokens.size() < attributes.width) continue;
+      tokens.erase(tokens.begin(),
+                   tokens.begin() + tokens.size() - attributes.width);
+      buffer.AddJoinedString(tokens, separator);
+      ++buffer_index;
+    }
+  }
+  output_row_splits[n_row_splits - 1] = buffer_index;
+
+  TfLiteTensor* output_values = GetOutput(context, node, kValues);
+  if (OutputIsTensor(node)) {
+    buffer.WriteToTensor(output_values, /*new_shape=*/nullptr);
+  } else {
+    buffer.WriteToTensorAsVector(output_values);
+  }
+
+  return kTfLiteOk;
+}
+
+}  // namespace ngrams
+
+TfLiteRegistration* Register_tftext_Ngrams() {
+  static TfLiteRegistration r = {ngrams::Init, ngrams::Free, ngrams::Prepare,
+                                 ngrams::Eval};
+  return &r;
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/ngrams.h b/tensorflow_lite_support/custom_ops/kernel/ngrams.h
new file mode 100644
index 0000000..5622906
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ngrams.h
@@ -0,0 +1,31 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_NGRAMS_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_NGRAMS_H_
+
+#include "tensorflow/lite/context.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+
+TfLiteRegistration* Register_tftext_Ngrams();
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_NGRAMS_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.cc b/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.cc
new file mode 100644
index 0000000..b87fcac
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.cc
@@ -0,0 +1,31 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.h"
+
+#include "tensorflow_lite_support/custom_ops/kernel/ngrams.h"
+#include "tensorflow/lite/mutable_op_resolver.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+
+void AddNgramsCustomOp(MutableOpResolver* resolver) {
+  resolver->AddCustom("tftext:Ngrams", Register_tftext_Ngrams());
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.h b/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.h
new file mode 100644
index 0000000..fc93268
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.h
@@ -0,0 +1,34 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_NGRAMS_OP_RESOLVER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_NGRAMS_OP_RESOLVER_H_
+
+#include "tensorflow/lite/mutable_op_resolver.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+
+// Adds the Ngrams custom op to an op resolver.
+// This function can be loaded using dlopen.  Since C++ function names get
+// mangled, declare this function as extern C, so its name is unchanged.
+extern "C" void AddNgramsCustomOp(MutableOpResolver* resolver);
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_NGRAMS_OP_RESOLVER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver_wrapper.cc b/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver_wrapper.cc
new file mode 100644
index 0000000..8274730
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver_wrapper.cc
@@ -0,0 +1,29 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "pybind11/pybind11.h"
+#include "tensorflow/lite/mutable_op_resolver.h"
+#include "tensorflow_lite_support/custom_ops/kernel/ngrams_op_resolver.h"
+
+PYBIND11_MODULE(_pywrap_ngrams_op_resolver, m) {
+  m.doc() = "_pywrap_ngrams_op_resolver";
+  m.def(
+      "AddNgramsCustomOp",
+      [](uintptr_t resolver) {
+        tflite::ops::custom::AddNgramsCustomOp(
+            reinterpret_cast<tflite::MutableOpResolver*>(resolver));
+      },
+      "Op registerer function for the tftext:Ngrams custom op.");
+}
diff --git a/tensorflow_lite_support/custom_ops/kernel/ngrams_test.cc b/tensorflow_lite_support/custom_ops/kernel/ngrams_test.cc
new file mode 100644
index 0000000..91ef47a
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ngrams_test.cc
@@ -0,0 +1,293 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/ngrams.h"
+
+#include <string>
+#include <vector>
+
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include "flatbuffers/flexbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/kernels/test_util.h"
+#include "tensorflow/lite/schema/schema_generated.h"
+#include "tensorflow/lite/string_util.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace ngrams {
+namespace test {
+namespace {
+
+using ::testing::ElementsAre;
+using ::testing::ElementsAreArray;
+
+}  // namespace
+
+class NgramsModel : public SingleOpModel {
+ public:
+  // Constructor for testing the op with a tf.Tensor
+  NgramsModel(int width, const std::string& string_separator,
+              const std::vector<std::string>& input_values,
+              const std::vector<int>& input_shape) {
+    input_values_ = AddInput(TensorType_STRING);
+    output_values_ = AddOutput(TensorType_STRING);
+
+    BuildCustomOp(width, string_separator);
+
+    BuildInterpreter({input_shape});
+    PopulateStringTensor(input_values_, input_values);
+    Invoke();
+  }
+
+  // Constructor for the op with a tf.RaggedTensor
+  // Note: This interface uses row_lengths, as they're closer to the
+  // dimensions in a TensorShape, but internally everything is row_splits.
+  NgramsModel(int width, const std::string& string_separator,
+              const std::vector<std::string>& input_values,
+              const std::vector<std::vector<int64_t>> nested_row_lengths) {
+    std::vector<std::vector<int>> input_shapes;
+    input_shapes.reserve(nested_row_lengths.size() + 1);
+
+    input_values_ = AddInput(TensorType_STRING);
+    input_shapes.push_back({static_cast<int>(input_values.size())});
+    output_values_ = AddOutput(TensorType_STRING);
+
+    input_row_splits_.reserve(nested_row_lengths.size());
+    output_row_splits_.reserve(nested_row_lengths.size());
+    for (int i = 0; i < nested_row_lengths.size(); ++i) {
+      input_row_splits_.push_back(AddInput(TensorType_INT64));
+      input_shapes.push_back(
+          {static_cast<int>(nested_row_lengths[i].size() + 1)});
+      output_row_splits_.push_back(AddOutput(TensorType_INT64));
+    }
+
+    BuildCustomOp(width, string_separator);
+
+    BuildInterpreter(input_shapes);
+    PopulateStringTensor(input_values_, input_values);
+    for (int i = 0; i < nested_row_lengths.size(); ++i) {
+      std::vector<int64_t> row_splits;
+      row_splits.reserve(nested_row_lengths[i].size() + 1);
+      int64_t index = 0;
+      row_splits.push_back(index);
+      for (int64_t row_length : nested_row_lengths[i]) {
+        index += row_length;
+        row_splits.push_back(index);
+      }
+      PopulateTensor(input_row_splits_[i], row_splits);
+    }
+    Invoke();
+  }
+
+  std::vector<int> GetValuesTensorShape() {
+    return GetTensorShape(output_values_);
+  }
+
+  std::vector<std::string> ExtractValuesTensorVector() {
+    std::vector<std::string> r;
+    TfLiteTensor* tensor = interpreter_->tensor(output_values_);
+    int n = GetStringCount(tensor);
+    for (int i = 0; i < n; ++i) {
+      StringRef ref = GetString(tensor, i);
+      r.emplace_back(ref.str, ref.len);
+    }
+    return r;
+  }
+
+  int GetNumNestedRowLengths() { return output_row_splits_.size(); }
+
+  std::vector<int> GetRowLengthsTensorShape(int i) {
+    std::vector<int> shape = GetTensorShape(output_row_splits_[i]);
+    --shape[0];
+    return shape;
+  }
+
+  std::vector<int64_t> ExtractRowLengthsTensorVector(int i) {
+    std::vector<int64_t> row_splits =
+        ExtractVector<int64_t>(output_row_splits_[i]);
+    std::vector<int64_t> row_lengths;
+    row_lengths.reserve(row_splits.size() - 1);
+    int64_t head = row_splits[0];
+    for (int i = 1; i < row_splits.size(); ++i) {
+      int64_t tail = row_splits[i];
+      row_lengths.push_back(tail - head);
+      head = tail;
+    }
+    return row_lengths;
+  }
+
+ private:
+  void BuildCustomOp(int width, const std::string& string_separator) {
+    flexbuffers::Builder fbb;
+    size_t start_map = fbb.StartMap();
+    fbb.Int("width", width);
+    fbb.String("string_separator", string_separator);
+    fbb.Int("axis", -1);
+    fbb.String("reduction_type", "STRING_JOIN");
+    fbb.EndMap(start_map);
+    fbb.Finish();
+
+    SetCustomOp("tftext:Ngrams", fbb.GetBuffer(), Register_tftext_Ngrams);
+  }
+
+  int input_values_;
+  std::vector<int> input_row_splits_;
+  int output_values_;
+  std::vector<int> output_row_splits_;
+};
+
+TEST(NgramsTest, TensorSingleSequenceWidthTwo) {
+  NgramsModel m(2, " ", {"this", "is", "a", "test"}, std::vector<int>{4});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(3));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this is", "is a", "a test"));
+}
+
+TEST(NgramsTest, TensorSingleSequenceWidthThree) {
+  NgramsModel m(3, " ", {"this", "is", "a", "test"}, std::vector<int>{4});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(2));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this is a", "is a test"));
+}
+
+TEST(NgramsTest, TensorSingleSequenceLongerSeparator) {
+  NgramsModel m(2, "...", {"this", "is", "a", "test"}, std::vector<int>{4});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(3));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this...is", "is...a", "a...test"));
+}
+
+TEST(NgramsTest, TensorSingleSequenceWidthTooLong) {
+  NgramsModel m(5, " ", {"this", "is", "a", "test"}, std::vector<int>{4});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(0));
+  EXPECT_THAT(m.ExtractValuesTensorVector(), ElementsAre());
+}
+
+TEST(NgramsTest, TensorMultidimensionalInputWidthTwo) {
+  NgramsModel m(2, " ",
+                {
+                    "0,0,0", "0,0,1", "0,0,2", "0,0,3",  //
+                    "0,1,0", "0,1,1", "0,1,2", "0,1,3",  //
+                    "0,2,0", "0,2,1", "0,2,2", "0,2,3",  //
+                    "1,0,0", "1,0,1", "1,0,2", "1,0,3",  //
+                    "1,1,0", "1,1,1", "1,1,2", "1,1,3",  //
+                    "1,2,0", "1,2,1", "1,2,2", "1,2,3",  //
+                },
+                std::vector<int>{2, 3, 4});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(2, 3, 3));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAreArray({
+                  "0,0,0 0,0,1", "0,0,1 0,0,2", "0,0,2 0,0,3",  //
+                  "0,1,0 0,1,1", "0,1,1 0,1,2", "0,1,2 0,1,3",  //
+                  "0,2,0 0,2,1", "0,2,1 0,2,2", "0,2,2 0,2,3",  //
+                  "1,0,0 1,0,1", "1,0,1 1,0,2", "1,0,2 1,0,3",  //
+                  "1,1,0 1,1,1", "1,1,1 1,1,2", "1,1,2 1,1,3",  //
+                  "1,2,0 1,2,1", "1,2,1 1,2,2", "1,2,2 1,2,3",  //
+              }));
+}
+
+TEST(NgramsTest, RaggedTensorSingleSequenceWidthTwo) {
+  std::vector<std::vector<int64_t>> nested_row_lengths;
+  nested_row_lengths.push_back({4});
+  NgramsModel m(2, " ", {"this", "is", "a", "test"},
+                nested_row_lengths);
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(3));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this is", "is a", "a test"));
+  ASSERT_THAT(m.GetNumNestedRowLengths(), 1);
+  EXPECT_THAT(m.GetRowLengthsTensorShape(0), ElementsAre(1));
+  EXPECT_THAT(m.ExtractRowLengthsTensorVector(0), ElementsAre(3));
+}
+
+TEST(NgramsTest, RaggedTensorSingleSequenceWidthThree) {
+  std::vector<std::vector<int64_t>> nested_row_lengths;
+  nested_row_lengths.push_back({4});
+  NgramsModel m(3, " ", {"this", "is", "a", "test"}, nested_row_lengths);
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(2));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this is a", "is a test"));
+  ASSERT_THAT(m.GetNumNestedRowLengths(), 1);
+  EXPECT_THAT(m.GetRowLengthsTensorShape(0), ElementsAre(1));
+  EXPECT_THAT(m.ExtractRowLengthsTensorVector(0), ElementsAre(2));
+}
+
+TEST(NgramsTest, RaggedTensorSingleSequenceLongerSeparator) {
+  std::vector<std::vector<int64_t>> nested_row_lengths;
+  nested_row_lengths.push_back({4});
+  NgramsModel m(2, "<>", {"this", "is", "a", "test"}, nested_row_lengths);
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(3));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this<>is", "is<>a", "a<>test"));
+  ASSERT_THAT(m.GetNumNestedRowLengths(), 1);
+  EXPECT_THAT(m.GetRowLengthsTensorShape(0), ElementsAre(1));
+  EXPECT_THAT(m.ExtractRowLengthsTensorVector(0), ElementsAre(3));
+}
+
+TEST(NgramsTest, RaggedTensorSingleSequenceWidthTooLong) {
+  std::vector<std::vector<int64_t>> nested_row_lengths;
+  nested_row_lengths.push_back({4});
+  NgramsModel m(5, " ", {"this", "is", "a", "test"}, nested_row_lengths);
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(0));
+  EXPECT_THAT(m.ExtractValuesTensorVector(), ElementsAre());
+  ASSERT_THAT(m.GetNumNestedRowLengths(), 1);
+  EXPECT_THAT(m.GetRowLengthsTensorShape(0), ElementsAre(1));
+  EXPECT_THAT(m.ExtractRowLengthsTensorVector(0), ElementsAre(0));
+}
+
+TEST(NgramsTest, RaggedTensorMultidimensionalInputWidthTwo) {
+  std::vector<std::vector<int64_t>> nested_row_lengths;
+  nested_row_lengths.push_back({4, 2, 1});
+  nested_row_lengths.push_back({5, 4, 3, 2, 2, 3, 4, 6});
+  NgramsModel m(2, " ",
+                {
+                    "0,0,0", "0,0,1", "0,0,2", "0,0,3", "0,0,4",           //
+                    "0,1,0", "0,1,1", "0,1,2", "0,1,3",                    //
+                    "0,2,0", "0,2,1", "0,2,2",                             //
+                    "0,3,0", "0,3,1",                                      //
+                    "1,0,0", "1,0,1",                                      //
+                    "1,1,0", "1,1,1", "1,1,2",                             //
+                    "1,2,0", "1,2,1", "1,2,2", "1,2,3",                    //
+                    "2,0,0", "2,0,1", "2,0,2", "2,0,3", "2,0,4", "2,0,5",  //
+                },
+                nested_row_lengths);
+
+  std::vector<std::string> expected_values = {
+      "0,0,0 0,0,1", "0,0,1 0,0,2", "0,0,2 0,0,3", "0,0,3 0,0,4",  //
+      "0,1,0 0,1,1", "0,1,1 0,1,2", "0,1,2 0,1,3",                 //
+      "0,2,0 0,2,1", "0,2,1 0,2,2",                                //
+      "0,3,0 0,3,1",                                               //
+      "1,0,0 1,0,1",                                               //
+      "1,1,0 1,1,1", "1,1,1 1,1,2",                                //
+      "1,2,0 1,2,1", "1,2,1 1,2,2", "1,2,2 1,2,3",                 //
+      "2,0,0 2,0,1", "2,0,1 2,0,2", "2,0,2 2,0,3", "2,0,3 2,0,4",
+      "2,0,4 2,0,5",  //
+  };
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(expected_values.size()));
+  EXPECT_THAT(m.ExtractValuesTensorVector(), ElementsAreArray(expected_values));
+  ASSERT_THAT(m.GetNumNestedRowLengths(), 2);
+  EXPECT_THAT(m.GetRowLengthsTensorShape(0), ElementsAre(3));
+  EXPECT_THAT(m.ExtractRowLengthsTensorVector(0), ElementsAre(4, 2, 1));
+  EXPECT_THAT(m.GetRowLengthsTensorShape(1), ElementsAre(8));
+  EXPECT_THAT(m.ExtractRowLengthsTensorVector(1),
+              ElementsAre(4, 3, 2, 1, 1, 2, 3, 5));
+}
+
+}  // namespace test
+}  // namespace ngrams
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/ngrams_test.py b/tensorflow_lite_support/custom_ops/kernel/ngrams_test.py
new file mode 100644
index 0000000..e52ca28
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ngrams_test.py
@@ -0,0 +1,266 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+# Lint as: python3
+"""Tests for tensorflow_lite_support.custom_ops.ngrams."""
+
+import os
+import sys
+import timeit
+
+from absl import logging
+from absl.testing import parameterized
+import tensorflow as tf
+import tensorflow_text as tf_text
+from tensorflow.lite.python import interpreter as interpreter_wrapper  # pylint: disable=g-direct-tensorflow-import
+from tensorflow_lite_support.custom_ops.python import tflite_text_api
+
+# Force loaded shared object symbols to be globally visible. This is needed so
+# that the interpreter_wrapper, in one .so file, can see the op resolver
+# in a different .so file. Note that this may already be set by default.
+# pylint: disable=g-import-not-at-top,g-bad-import-order,unused-import
+if hasattr(sys, 'setdlopenflags') and hasattr(sys, 'getdlopenflags'):
+  sys.setdlopenflags(sys.getdlopenflags() | os.RTLD_GLOBAL)
+from tensorflow_lite_support.custom_ops.kernel import _pywrap_ngrams_op_resolver
+
+TEST_CASES = [
+    [['this', 'is', 'a', 'test']],
+    [['one']],
+    [['two', 'tokens'], ['a', 'b']],
+    [['has', 'three', 'tokens'], ['a', 'b', 'c'], ['0', '1', '2']],
+    [['a', 'ragged', 'tensor'], ['a'], ['0', '1']],
+    [[['a', 'multidimensional', 'test', 'case'], ['a', 'b', 'c', 'd', 'e']],
+     [['0', '1', '2', '3', '4', '5']]],
+]
+
+INVOKES_FOR_SINGLE_OP_BENCHMARK = 1000
+INVOKES_FOR_FLEX_DELEGATE_BENCHMARK = 100
+
+
+class NgramsTest(parameterized.TestCase):
+
+  _models = {}
+
+  def _make_model(self, rank, width, ragged_tensor=False, flex=False):
+    temp_dir = self.create_tempdir().full_path
+
+    key = (rank, width, ragged_tensor, flex)
+    if key in self._models:
+      return self._models[key]
+
+    ngrams = tf_text.ngrams if flex else tflite_text_api.ngrams
+
+    if ragged_tensor:
+      input_signature = [tf.TensorSpec(shape=[None], dtype=tf.string)]
+      rs = rank - 1
+      input_signature += [tf.TensorSpec(shape=[None], dtype=tf.int64)] * rs
+
+      class Model(tf.Module):
+
+        @tf.function(input_signature=input_signature)
+        def __call__(self, values, *args):
+          row_splits = list(args)
+          row_splits.reverse()
+          input_tensor = tf.RaggedTensor.from_nested_row_splits(
+              flat_values=values, nested_row_splits=tuple(row_splits))
+          output_tensor = ngrams(
+              input_tensor, width, reduction_type=tf_text.Reduction.STRING_JOIN)
+          output = [output_tensor.flat_values]
+          output.extend(list(output_tensor.nested_row_splits))
+          output.reverse()
+          return tuple(output)
+
+      tf.saved_model.save(Model(), temp_dir)
+    else:
+      shape = [None] * rank
+
+      class Model(tf.Module):
+
+        @tf.function(
+            input_signature=[tf.TensorSpec(shape=shape, dtype=tf.string)])
+        def __call__(self, input_tensor):
+          return ngrams(
+              input_tensor, width, reduction_type=tf_text.Reduction.STRING_JOIN)
+
+      tf.saved_model.save(Model(), temp_dir)
+
+    converter = tf.lite.TFLiteConverter.from_saved_model(temp_dir)
+    converter.inference_type = tf.float32
+    converter.inference_input_type = tf.float32
+    converter.allow_custom_ops = not flex
+    if flex:
+      converter.target_spec.supported_ops = [
+          tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS
+      ]
+    model = converter.convert()
+    self._models[key] = model
+    return model
+
+  @parameterized.parameters([t] for t in TEST_CASES)
+  def test_width_2_tensor_equivalence(self, test_case):
+    input_tensor = tf.ragged.constant(test_case).to_tensor()
+    tf_output = tf_text.ngrams(
+        input_tensor, 2, reduction_type=tf_text.Reduction.STRING_JOIN)
+
+    rank = input_tensor.shape.rank
+    model = self._make_model(rank, 2, ragged_tensor=False, flex=False)
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=model, custom_op_registerers=['AddNgramsCustomOp'])
+    interpreter.resize_tensor_input(0, input_tensor.shape)
+    interpreter.allocate_tensors()
+    interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                           input_tensor.numpy())
+    interpreter.invoke()
+    tflite_output = interpreter.get_tensor(
+        interpreter.get_output_details()[0]['index'])
+
+    self.assertEqual(tf_output.numpy().tolist(), tflite_output.tolist())
+
+  @parameterized.parameters([t] for t in TEST_CASES)
+  def test_width_3_tensor_equivalence(self, test_case):
+    input_tensor = tf.ragged.constant(test_case).to_tensor()
+    tf_output = tf_text.ngrams(
+        input_tensor, 3, reduction_type=tf_text.Reduction.STRING_JOIN)
+
+    rank = input_tensor.shape.rank
+    model = self._make_model(rank, 3, ragged_tensor=False, flex=False)
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=model, custom_op_registerers=['AddNgramsCustomOp'])
+    interpreter.resize_tensor_input(0, input_tensor.shape)
+    interpreter.allocate_tensors()
+    interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                           input_tensor.numpy())
+    interpreter.invoke()
+    tflite_output = interpreter.get_tensor(
+        interpreter.get_output_details()[0]['index'])
+    self.assertEqual(tf_output.numpy().tolist(), tflite_output.tolist())
+
+  @parameterized.parameters([t] for t in TEST_CASES)
+  def test_width_2_ragged_tensor_equivalence(self, test_case):
+    input_tensor = tf.ragged.constant(test_case)
+    tf_output = tf_text.ngrams(
+        input_tensor, 2, reduction_type=tf_text.Reduction.STRING_JOIN)
+
+    rank = input_tensor.shape.rank
+    model = self._make_model(rank, 2, ragged_tensor=True, flex=False)
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=model, custom_op_registerers=['AddNgramsCustomOp'])
+    interpreter.resize_tensor_input(0, input_tensor.flat_values.shape)
+    for r in range(rank - 1):
+      interpreter.resize_tensor_input(r + 1,
+                                      input_tensor.nested_row_splits[r].shape)
+    interpreter.allocate_tensors()
+    interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                           input_tensor.flat_values.numpy())
+    for r in range(rank - 1):
+      interpreter.set_tensor(interpreter.get_input_details()[r + 1]['index'],
+                             input_tensor.nested_row_splits[r].numpy())
+    interpreter.invoke()
+    tflite_output_values = interpreter.get_tensor(
+        interpreter.get_output_details()[0]['index'])
+    self.assertEqual(tf_output.flat_values.numpy().tolist(),
+                     tflite_output_values.tolist())
+    for i in range(rank - 1):
+      tflite_output_cur_row_splits = interpreter.get_tensor(
+          interpreter.get_output_details()[i + 1]['index'])
+      self.assertEqual(tf_output.nested_row_splits[i].numpy().tolist(),
+                       tflite_output_cur_row_splits.tolist())
+
+  @parameterized.parameters([t] for t in TEST_CASES)
+  def test_width_3_ragged_tensor_equivalence(self, test_case):
+    input_tensor = tf.ragged.constant(test_case)
+    tf_output = tf_text.ngrams(
+        input_tensor, 3, reduction_type=tf_text.Reduction.STRING_JOIN)
+
+    rank = input_tensor.shape.rank
+    model = self._make_model(rank, 3, ragged_tensor=True, flex=False)
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=model, custom_op_registerers=['AddNgramsCustomOp'])
+    interpreter.resize_tensor_input(0, input_tensor.flat_values.shape)
+    for r in range(rank - 1):
+      interpreter.resize_tensor_input(r + 1,
+                                      input_tensor.nested_row_splits[r].shape)
+    interpreter.allocate_tensors()
+    interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                           input_tensor.flat_values.numpy())
+    for r in range(rank - 1):
+      interpreter.set_tensor(interpreter.get_input_details()[r + 1]['index'],
+                             input_tensor.nested_row_splits[r].numpy())
+    interpreter.invoke()
+    tflite_output_values = interpreter.get_tensor(
+        interpreter.get_output_details()[0]['index'])
+    self.assertEqual(tf_output.flat_values.numpy().tolist(),
+                     tflite_output_values.tolist())
+    for i in range(rank - 1):
+      tflite_output_cur_row_splits = interpreter.get_tensor(
+          interpreter.get_output_details()[i + 1]['index'])
+      self.assertEqual(tf_output.nested_row_splits[i].numpy().tolist(),
+                       tflite_output_cur_row_splits.tolist())
+
+  def test_latency(self):
+    latency_op = 0.0
+    for test_case in TEST_CASES:
+      input_tensor = tf.ragged.constant(test_case)
+
+      rank = input_tensor.shape.rank
+      model = self._make_model(rank, 3, ragged_tensor=True, flex=False)
+      interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+          model_content=model, custom_op_registerers=['AddNgramsCustomOp'])
+      interpreter.resize_tensor_input(0, input_tensor.flat_values.shape)
+      for r in range(rank - 1):
+        interpreter.resize_tensor_input(r + 1,
+                                        input_tensor.nested_row_splits[r].shape)
+      interpreter.allocate_tensors()
+      interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                             input_tensor.flat_values.numpy())
+      for r in range(rank - 1):
+        interpreter.set_tensor(interpreter.get_input_details()[r + 1]['index'],
+                               input_tensor.nested_row_splits[r].numpy())
+      start_time = timeit.default_timer()
+      for _ in range(INVOKES_FOR_SINGLE_OP_BENCHMARK):
+        interpreter.invoke()
+      latency_op = latency_op + timeit.default_timer() - start_time
+    latency_op = latency_op / (
+        INVOKES_FOR_SINGLE_OP_BENCHMARK * len(TEST_CASES))
+
+    latency_flex = 0.0
+    for test_case in TEST_CASES:
+      input_tensor = tf.ragged.constant(test_case)
+
+      rank = input_tensor.shape.rank
+      model = self._make_model(rank, 3, ragged_tensor=True, flex=True)
+      interpreter = interpreter_wrapper.Interpreter(model_content=model)
+      interpreter.resize_tensor_input(0, input_tensor.flat_values.shape)
+      for r in range(rank - 1):
+        interpreter.resize_tensor_input(r + 1,
+                                        input_tensor.nested_row_splits[r].shape)
+      interpreter.allocate_tensors()
+      interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                             input_tensor.flat_values.numpy())
+      for r in range(rank - 1):
+        interpreter.set_tensor(interpreter.get_input_details()[r + 1]['index'],
+                               input_tensor.nested_row_splits[r].numpy())
+      start_time = timeit.default_timer()
+      for _ in range(INVOKES_FOR_FLEX_DELEGATE_BENCHMARK):
+        interpreter.invoke()
+      latency_flex = latency_flex + timeit.default_timer() - start_time
+    latency_flex = latency_flex / (
+        INVOKES_FOR_FLEX_DELEGATE_BENCHMARK * len(TEST_CASES))
+
+    logging.info('Latency (single op): %fms', latency_op * 1000.0)
+    logging.info('Latency (flex delegate): %fms', latency_flex * 1000.0)
+
+
+if __name__ == '__main__':
+  tf.test.main()
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/BUILD b/tensorflow_lite_support/custom_ops/kernel/ragged/BUILD
new file mode 100644
index 0000000..a512cdc
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/BUILD
@@ -0,0 +1,81 @@
+# RaggedTensors suppport in TFLite
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "ragged_tensor_to_tensor_tflite",
+    srcs = ["ragged_tensor_to_tensor_tflite.cc"],
+    visibility = [
+        "//visibility:public",
+    ],
+    deps = [
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/core/util:ragged_to_dense_util_common",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:types",
+    ],
+)
+
+cc_test(
+    name = "ragged_tensor_to_tensor_tflite_test",
+    srcs = ["ragged_tensor_to_tensor_tflite_test.cc"],
+    deps = [
+        ":ragged_tensor_to_tensor_tflite",
+        "@com_google_googletest//:gtest_main",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/kernels:test_util",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+        "@org_tensorflow//tensorflow/lite/schema:schema_fbs",
+    ],
+)
+
+cc_library(
+    name = "py_tflite_registerer",
+    srcs = ["py_tflite_registerer.cc"],
+    hdrs = ["py_tflite_registerer.h"],
+    deps = [
+        ":ragged_tensor_to_tensor_tflite",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+    ],
+    alwayslink = 1,
+)
+
+cc_library(
+    name = "ragged_range_tflite",
+    srcs = ["ragged_range_tflite.cc"],
+    visibility = [
+        "//visibility:public",
+    ],
+    deps = [
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:types",
+    ],
+)
+
+cc_test(
+    name = "ragged_range_tflite_test",
+    srcs = ["ragged_range_tflite_test.cc"],
+    deps = [
+        ":ragged_range_tflite",
+        "@com_google_googletest//:gtest_main",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/kernels:test_util",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+        "@org_tensorflow//tensorflow/lite/schema:schema_fbs",
+    ],
+)
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/py/BUILD b/tensorflow_lite_support/custom_ops/kernel/ragged/py/BUILD
new file mode 100644
index 0000000..650ab90
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/py/BUILD
@@ -0,0 +1,27 @@
+# Python wrapper used for test.
+
+load("@org_tensorflow//tensorflow:tensorflow.bzl", "pybind_extension")
+
+package(
+    default_visibility = [
+        "//tensorflow_lite_support:users",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+pybind_extension(
+    name = "pywrap_tflite_registerer",
+    srcs = [
+        "pywrap_tflite_registerer.cc",
+    ],
+    additional_exported_symbols = ["TFLite_RaggedTensorToTensorRegisterer"],
+    module_name = "pywrap_tflite_registerer",
+    srcs_version = "PY3ONLY",
+    deps = [
+        "//tensorflow_lite_support/custom_ops/kernel/ragged:py_tflite_registerer",
+        "@local_config_python//:python_headers",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+        "@pybind11",
+    ],
+)
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/py/pywrap_tflite_registerer.cc b/tensorflow_lite_support/custom_ops/kernel/ragged/py/pywrap_tflite_registerer.cc
new file mode 100644
index 0000000..0b9432a
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/py/pywrap_tflite_registerer.cc
@@ -0,0 +1,35 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "pybind11/pybind11.h"
+#include "pybind11/pytypes.h"
+#include "tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.h"
+
+PYBIND11_MODULE(pywrap_tflite_registerer, m) {
+  m.doc() = R"pbdoc(
+    pywrap_tflite_registerer
+    A module with a wrapper that adds to a Python wrapper for TFLite
+    ragged_tensor_to_tensor.
+  )pbdoc";
+  m.def(
+      "TFLite_RaggedTensorToTensorRegisterer",
+      [](uintptr_t resolver) {
+        TFLite_RaggedTensorToTensorRegisterer(
+            reinterpret_cast<tflite::MutableOpResolver*>(resolver));
+      },
+      R"pbdoc(
+      The function that adds RaggedTensorToTensor to the TFLite interpreter.
+      )pbdoc");
+}
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.cc b/tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.cc
new file mode 100644
index 0000000..7c93d8b
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.cc
@@ -0,0 +1,31 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.h"
+
+#include "tensorflow/lite/mutable_op_resolver.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+TfLiteRegistration* Register_RAGGED_TENSOR_TO_TENSOR();
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+extern "C" void TFLite_RaggedTensorToTensorRegisterer(
+    tflite::MutableOpResolver* resolver) {
+  resolver->AddCustom("RaggedTensorToTensor",
+                      tflite::ops::custom::Register_RAGGED_TENSOR_TO_TENSOR());
+}
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.h b/tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.h
new file mode 100644
index 0000000..ade3c5c
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/py_tflite_registerer.h
@@ -0,0 +1,25 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_RAGGED_PY_TFLITE_REGISTERER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_RAGGED_PY_TFLITE_REGISTERER_H_
+
+#include "tensorflow/lite/mutable_op_resolver.h"
+
+// C-function that is called from the Python Wrapper.
+
+extern "C" void TFLite_RaggedTensorToTensorRegisterer(
+    tflite::MutableOpResolver *resolver);
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_RAGGED_PY_TFLITE_REGISTERER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_range_tflite.cc b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_range_tflite.cc
new file mode 100644
index 0000000..a35a6db
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_range_tflite.cc
@@ -0,0 +1,192 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include <algorithm>
+#include <functional>
+
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/context.h"
+#include "tensorflow/lite/kernels/internal/tensor.h"
+#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
+#include "tensorflow/lite/kernels/internal/types.h"
+#include "tensorflow/lite/kernels/kernel_util.h"
+#include "tensorflow/lite/model.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace ragged {
+namespace ragged_range {
+namespace {
+constexpr int kInputStarts = 0;
+constexpr int kInputLimits = 1;
+constexpr int kInputDeltas = 2;
+
+constexpr int kOutputNestedSplits = 0;
+constexpr int kOutputDenseValues = 1;
+
+TfLiteIntArray* IntArrayFromInt(int x) {
+  TfLiteIntArray* result = TfLiteIntArrayCreate(1);
+  result->data[0] = x;
+  return result;
+}
+
+// Returns the number of elements in the specified range.
+template <typename T, typename SPLITS_TYPE>
+SPLITS_TYPE RangeSize(T start, T limit, T delta) {
+  if (((delta > 0) && (limit < start)) || ((delta < 0) && (limit > start))) {
+    return 0;
+  }
+  // The following is copied from tensorflow::RangeOp::Compute().
+  return (
+      std::is_integral<T>::value
+          ? ((std::abs(limit - start) + std::abs(delta) - 1) / std::abs(delta))
+          : std::ceil(std::abs((limit - start) / delta)));
+}
+
+template <typename T, typename SPLITS_TYPE>
+TfLiteStatus EvalT(TfLiteContext* context, TfLiteNode* node) {
+  TfLiteTensor& input_starts =
+      context->tensors[node->inputs->data[kInputStarts]];
+  TfLiteTensor& input_limits =
+      context->tensors[node->inputs->data[kInputLimits]];
+  TfLiteTensor& input_deltas =
+      context->tensors[node->inputs->data[kInputDeltas]];
+  // Determine which tensors we need to broadcast.
+  const bool broadcast_starts = NumElements(&input_starts) == 1;
+  const bool broadcast_limits = NumElements(&input_limits) == 1;
+  const bool broadcast_deltas = NumElements(&input_deltas) == 1;
+
+  // nrows (number of output rows) is the size of the non-broadcast inputs,
+  // or 1 if all inputs are scalars.
+  std::vector<int> in_sizes;
+  if (!broadcast_starts) in_sizes.push_back(input_starts.dims->data[0]);
+  if (!broadcast_limits) in_sizes.push_back(input_limits.dims->data[0]);
+  if (!broadcast_deltas) in_sizes.push_back(input_deltas.dims->data[0]);
+  if (std::adjacent_find(std::begin(in_sizes), std::end(in_sizes),
+                         std::not_equal_to<>()) != std::end(in_sizes)) {
+    context->ReportError(
+        context,
+        "Invalid argument: starts, limits, and deltas must have the "
+        "same shape");
+    return kTfLiteError;
+  }
+
+  const SPLITS_TYPE nrows = in_sizes.empty() ? 1 : in_sizes.front();
+
+  const T* starts = GetTensorData<T>(&input_starts);
+  const T* limits = GetTensorData<T>(&input_limits);
+  const T* deltas = GetTensorData<T>(&input_deltas);
+
+  TfLiteTensor& rt_nested_splits_out =
+      context->tensors[node->outputs->data[kOutputNestedSplits]];
+  TF_LITE_ENSURE_OK(context,
+                    context->ResizeTensor(context, &rt_nested_splits_out,
+                                          IntArrayFromInt(nrows + 1)));
+  SPLITS_TYPE* rt_nested_splits =
+      GetTensorData<SPLITS_TYPE>(&rt_nested_splits_out);
+  rt_nested_splits[0] = 0;
+
+  for (int row = 0; row < nrows; ++row) {
+    const T start = broadcast_starts ? starts[0] : starts[row];
+    const T limit = broadcast_limits ? limits[0] : limits[row];
+    const T delta = broadcast_deltas ? deltas[0] : deltas[row];
+    if (delta == 0) {
+      context->ReportError(context, "Invalid argument: Requires delta != 0");
+      return kTfLiteError;
+    }
+    rt_nested_splits[row + 1] =
+        rt_nested_splits[row] + RangeSize<T, SPLITS_TYPE>(start, limit, delta);
+  }
+  const SPLITS_TYPE nvals = rt_nested_splits[nrows];
+
+  TfLiteTensor& rt_dense_values_out =
+      context->tensors[node->outputs->data[kOutputDenseValues]];
+  TF_LITE_ENSURE_OK(context,
+                    context->ResizeTensor(context, &rt_dense_values_out,
+                                          IntArrayFromInt(nvals)));
+  T* rt_dense_values = GetTensorData<T>(&rt_dense_values_out);
+  int value_index = 0;
+  for (int row = 0; row < nrows; ++row) {
+    const SPLITS_TYPE row_size =
+        rt_nested_splits[row + 1] - rt_nested_splits[row];
+    T value = broadcast_starts ? starts[0] : starts[row];
+    const T delta = broadcast_deltas ? deltas[0] : deltas[row];
+    for (SPLITS_TYPE i = 0; i < row_size; ++i) {
+      rt_dense_values[value_index++] = value;
+      value += delta;
+    }
+  }
+  return kTfLiteOk;
+}
+
+template <typename SPLITS_TYPE>
+TfLiteStatus EvalSplitsT(TfLiteContext* context, TfLiteNode* node) {
+  TfLiteTensor& rt_dense_values_out =
+      context->tensors[node->outputs->data[kOutputDenseValues]];
+  switch (rt_dense_values_out.type) {
+    case kTfLiteInt32:
+      return EvalT<int32_t, SPLITS_TYPE>(context, node);
+    case kTfLiteInt64:
+      return EvalT<int64_t, SPLITS_TYPE>(context, node);
+    case kTfLiteFloat32:
+      return EvalT<float, SPLITS_TYPE>(context, node);
+    case kTfLiteFloat64:
+      return EvalT<double, SPLITS_TYPE>(context, node);
+    default:
+      context->ReportError(context,
+                           "Invalid argument: Not supported VALUES type");
+      return kTfLiteError;
+  }
+}
+}  // namespace
+
+TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
+  // Set outputs dynamic.
+  TfLiteTensor& nested_splits =
+      context->tensors[node->outputs->data[kOutputNestedSplits]];
+  SetTensorToDynamic(&nested_splits);
+  TfLiteTensor& dense_values =
+      context->tensors[node->outputs->data[kOutputDenseValues]];
+  SetTensorToDynamic(&dense_values);
+  return kTfLiteOk;
+}
+
+TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
+  TfLiteTensor& rt_nested_splits_out =
+      context->tensors[node->outputs->data[kOutputNestedSplits]];
+  switch (rt_nested_splits_out.type) {
+    case kTfLiteInt32:
+      return EvalSplitsT<int32_t>(context, node);
+    case kTfLiteInt64:
+      return EvalSplitsT<int64_t>(context, node);
+    default:
+      context->ReportError(context,
+                           "Invalid argument: Not supported ROW_SPLITS type");
+      return kTfLiteError;
+  }
+}
+
+}  // namespace ragged_range
+}  // namespace ragged
+TfLiteRegistration* Register_RAGGED_RANGE() {
+  static TfLiteRegistration r = {nullptr /*Initialize*/, nullptr /*Free*/,
+                                 ragged::ragged_range::Prepare,
+                                 ragged::ragged_range::Eval};
+  return &r;
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_range_tflite_test.cc b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_range_tflite_test.cc
new file mode 100644
index 0000000..54cf445
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_range_tflite_test.cc
@@ -0,0 +1,155 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include <initializer_list>
+#include <vector>
+
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include "flatbuffers/flexbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/interpreter.h"
+#include "tensorflow/lite/kernels/internal/tensor.h"
+#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
+#include "tensorflow/lite/kernels/test_util.h"
+#include "tensorflow/lite/schema/schema_generated.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+TfLiteRegistration* Register_RAGGED_RANGE();
+}  // namespace custom
+}  // namespace ops
+
+namespace {
+
+template <typename T>
+class RaggedRangeOpModel : public SingleOpModel {
+ public:
+  static TensorType GetType();
+
+  RaggedRangeOpModel(const std::vector<T>& start, const std::vector<T>& limits,
+                     const std::vector<T>& deltas) {
+    const TensorType value_type = GetType();
+    std::vector<std::vector<int>> shapes;
+    input_start_ = AddInput(value_type);
+    shapes.push_back({static_cast<int>(start.size())});
+    input_limits_ = AddInput(value_type);
+    shapes.push_back({static_cast<int>(limits.size())});
+    input_deltas_ = AddInput(value_type);
+    shapes.push_back({static_cast<int>(deltas.size())});
+
+    output_splits_ = AddOutput(TensorType_INT32);
+    output_values_ = AddOutput(value_type);
+
+    SetCustomOp("RaggedRange", {}, ops::custom::Register_RAGGED_RANGE);
+    BuildInterpreter(shapes);
+
+    PopulateTensor(input_start_, start);
+    PopulateTensor(input_limits_, limits);
+    PopulateTensor(input_deltas_, deltas);
+  }
+
+  std::vector<int32> GetSplits() {
+    return ExtractVector<int32>(output_splits_);
+  }
+  std::vector<T> GetValues() const { return ExtractVector<T>(output_values_); }
+
+ protected:
+  int input_start_ = -1;
+  int input_limits_ = -1;
+  int input_deltas_ = -1;
+
+  int output_splits_ = -1;
+  int output_values_ = -1;
+};
+
+template <>
+TensorType RaggedRangeOpModel<int32>::GetType() {
+  return TensorType_INT32;
+}
+
+template <>
+TensorType RaggedRangeOpModel<float>::GetType() {
+  return TensorType_FLOAT32;
+}
+
+TEST(RaggedRangeOpTest, IntValues) {
+  RaggedRangeOpModel<int32> model({0, 5, 8, 5},    // Starts.
+                                  {8, 7, 8, 1},    // Limits.
+                                  {2, 1, 1, -1});  // Deltas.
+  model.Invoke();
+
+  EXPECT_THAT(model.GetSplits(),
+              testing::UnorderedElementsAreArray({0, 4, 6, 6, 10}));
+  EXPECT_THAT(model.GetValues(), testing::UnorderedElementsAreArray(
+                                     {0, 2, 4, 6, 5, 6, 5, 4, 3, 2}));
+}
+
+TEST(RaggedRangeOpTest, FloatValues) {
+  RaggedRangeOpModel<float> model({0, 5, 8, 5},    // Starts.
+                                  {8, 7, 8, 1},    // Limits.
+                                  {2, 1, 1, -1});  // Deltas.
+  model.Invoke();
+
+  EXPECT_THAT(model.GetSplits(),
+              testing::UnorderedElementsAreArray({0, 4, 6, 6, 10}));
+  EXPECT_THAT(model.GetValues(), testing::UnorderedElementsAreArray(
+                                     {0, 2, 4, 6, 5, 6, 5, 4, 3, 2}));
+}
+
+TEST(RaggedRangeOpTest, BroadcastDelta) {
+  RaggedRangeOpModel<int32> model({0, 5, 8},  // Starts.
+                                  {8, 7, 8},  // Limits.
+                                  {1});       // Deltas.
+  model.Invoke();
+
+  EXPECT_THAT(model.GetSplits(),
+              testing::UnorderedElementsAreArray({0, 8, 10, 10}));
+  EXPECT_THAT(model.GetValues(), testing::UnorderedElementsAreArray(
+                                     {0, 1, 2, 3, 4, 5, 6, 7, 5, 6}));
+}
+
+TEST(RaggedRangeOpTest, BroadcastStartDeltas) {
+  RaggedRangeOpModel<int32> model({0},      // Starts.
+                                  {10},     // Limits.
+                                  {2, 1});  // Deltas.
+  model.Invoke();
+
+  EXPECT_THAT(model.GetSplits(),
+              testing::UnorderedElementsAreArray({0, 5, 15}));
+  EXPECT_THAT(model.GetValues(),
+              testing::UnorderedElementsAreArray(
+                  {0, 2, 4, 6, 8, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9}));
+}
+
+TEST(RaggedRangeOpTest, BadDeltas) {
+  RaggedRangeOpModel<int32> model({0, 5, 8, 5},   // Starts.
+                                  {8, 7, 7, 9},   // Limits.
+                                  {0, 1, 1, 1});  // Deltas.
+  EXPECT_EQ(model.InvokeUnchecked(), kTfLiteError);
+}
+
+TEST(RaggedRangeOpTest, ZeroRange) {
+  RaggedRangeOpModel<int32> model({0, 7},   // Starts.
+                                  {8, 5},   // Limits.
+                                  {1, 1});  // Deltas.
+  model.Invoke();
+  EXPECT_THAT(model.GetSplits(), testing::UnorderedElementsAreArray({0, 8, 8}));
+  EXPECT_THAT(model.GetValues(),
+              testing::UnorderedElementsAreArray({0, 1, 2, 3, 4, 5, 6, 7}));
+}
+
+}  // namespace
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_tensor_to_tensor_tflite.cc b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_tensor_to_tensor_tflite.cc
new file mode 100644
index 0000000..09ac76c
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_tensor_to_tensor_tflite.cc
@@ -0,0 +1,690 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <cstdint>
+#include <memory>
+
+#include "flatbuffers/flexbuffers.h"  // from @flatbuffers
+#include "tensorflow/core/util/ragged_to_dense_util_common.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/context.h"
+#include "tensorflow/lite/kernels/internal/tensor.h"
+#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
+#include "tensorflow/lite/kernels/internal/types.h"
+#include "tensorflow/lite/kernels/kernel_util.h"
+#include "tensorflow/lite/model.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace ragged {
+namespace ragged_tensor_to_tensor {
+namespace {
+
+constexpr int kShapeInput = 0;
+constexpr int kValuesInput = 1;
+constexpr int kDefaultValueInput = 2;
+constexpr int kFirstPartitionInputIndex = 3;
+
+constexpr int kOutputTensor = 0;
+
+constexpr char kRowPartitionTypesAttr[] = "row_partition_types";
+
+struct ConversionAttributes {
+  std::vector<tensorflow::RowPartitionType> partition_types;
+  int ragged_rank = 0;
+
+  tensorflow::RowPartitionType GetRowPartitionTypeByDimension(
+      int dimension) const {
+    if (partition_types.front() ==
+        tensorflow::RowPartitionType::FIRST_DIM_SIZE) {
+      return partition_types[dimension + 1];
+    } else {
+      return partition_types[dimension];
+    }
+  }
+};
+template <typename INDEX_TYPE>
+int GetFirstDimensionSizeT(TfLiteContext* context,
+                           const TfLiteTensor& first_partition_input,
+                           const ConversionAttributes* attributes) {
+  const tensorflow::RowPartitionType first_partition_type =
+      attributes->partition_types.front();
+  switch (first_partition_type) {
+    case tensorflow::RowPartitionType::FIRST_DIM_SIZE:
+      return *GetTensorData<INDEX_TYPE>(&first_partition_input);
+    case tensorflow::RowPartitionType::VALUE_ROWIDS:
+      context->ReportError(context,
+                           "Cannot handle VALUE_ROWIDS in first dimension.");
+      return -1;
+    case tensorflow::RowPartitionType::ROW_SPLITS: {
+      const auto shape = GetTensorShape(&first_partition_input);
+      return shape.Dims(0) - 1;
+    }
+
+    default:
+      context->ReportError(
+          context, "Cannot handle type ",
+          RowPartitionTypeToString(first_partition_type).c_str());
+      return -1;
+  }
+}
+
+int GetFirstDimensionSize(TfLiteContext* context,
+                          const TfLiteTensor& first_partition_input,
+                          const ConversionAttributes* attributes) {
+  switch (first_partition_input.type) {
+    case kTfLiteInt32:
+      return GetFirstDimensionSizeT<int32_t>(context, first_partition_input,
+                                             attributes);
+    case kTfLiteInt64:
+      return GetFirstDimensionSizeT<int64_t>(context, first_partition_input,
+                                             attributes);
+    default:
+      context->ReportError(context,
+                           "Not supported row partitioning tensor type");
+      return -1;
+  }
+}
+
+bool ValidateDefaultValueShape(TfLiteContext* context,
+                               const RuntimeShape& default_value_shape,
+                               const RuntimeShape& /*value_shape*/) {
+  // TF implementation also checks that shapes are not defined, not needed in
+  // TFLite.
+  // TODO(mgubin): Only scalar default value sizes are supported.
+  if (default_value_shape.FlatSize() != 1) {
+    context->ReportError(context, "Only scalar default value is supported");
+    return false;
+  }
+  return true;
+}
+
+RuntimeShape TensorShapeFromTensor(const TfLiteTensor& tensor) {
+  // TODO(mgubin): No checks, see
+  // third_party/tensorflow/core/kernels/list_kernels.cc
+  const RuntimeShape tensor_shape(tensor.dims->size, tensor.dims->data);
+  if (0 == tensor.dims->size) {
+    // If the input tensor is scalar then the shape is empty (also scalar).
+    return RuntimeShape{};
+  }
+  RuntimeShape result(tensor_shape.FlatSize());
+  switch (tensor.type) {
+    case kTfLiteInt32: {
+      for (int i = 0; i < tensor_shape.FlatSize(); ++i) {
+        result.SetDim(i, GetTensorData<int32_t>(&tensor)[i]);
+      }
+    } break;
+    case kTfLiteInt64: {
+      for (int i = 0; i < tensor_shape.FlatSize(); ++i) {
+        result.SetDim(i, GetTensorData<int64_t>(&tensor)[i]);
+      }
+    } break;
+    default: {
+      // Checked in Prepare.
+    }
+  }
+  return result;
+}
+
+const TfLiteTensor* GetRowPartitionTensor(
+    const ConversionAttributes& conversion_attributes, TfLiteContext* context,
+    TfLiteNode* node, int dimension) {
+  if (conversion_attributes.partition_types.front() ==
+      tensorflow::RowPartitionType::FIRST_DIM_SIZE) {
+    return &context->tensors[node->inputs->data[kFirstPartitionInputIndex + 1 +
+                                                dimension]];
+  } else {
+    return &context->tensors[node->inputs
+                                 ->data[kFirstPartitionInputIndex + dimension]];
+  }
+}
+
+int GetMaxWidthValueRowID(const TfLiteTensor* tensor) {
+  const RuntimeShape tensor_shape(tensor->dims->size, tensor->dims->data);
+  const int index_length = tensor_shape.FlatSize();
+  if (index_length == 0) {
+    return 0;
+  }
+  auto value_rowids = [tensor](int index) {
+    switch (tensor->type) {
+      case kTfLiteInt32:
+        return static_cast<int>(tensor->data.i32[index]);
+      case kTfLiteInt64:
+        return static_cast<int>(tensor->data.i64[index]);
+      default:
+        // TODO(mgubin): Add error checks.
+        return 0;
+    }
+  };
+  int first_equal_index = 0;
+  int first_equal_index_value = value_rowids(0);
+  int max_width = 0;
+  for (int i = 0; i < index_length; ++i) {
+    const int value = value_rowids(i);
+    if (value != first_equal_index_value) {
+      first_equal_index_value = value;
+      max_width = std::max(i - first_equal_index, max_width);
+      first_equal_index = i;
+    }
+  }
+  return std::max(index_length - first_equal_index, max_width);
+}
+
+int GetMaxWidthRowSplit(const TfLiteTensor* tensor) {
+  const RuntimeShape tensor_shape(tensor->dims->size, tensor->dims->data);
+  const int tensor_length = tensor_shape.FlatSize();
+  if (tensor_length == 0 || tensor_length == 1) {
+    return 0;
+  }
+  auto value_rowsplit = [tensor](int index) {
+    switch (tensor->type) {
+      case kTfLiteInt32:
+        return static_cast<int>(tensor->data.i32[index]);
+      case kTfLiteInt64:
+        return static_cast<int>(tensor->data.i64[index]);
+      default:
+        // TODO(mgubin): Add error checks.
+        return 0;
+    }
+  };
+  int max_width = 1;
+  int prev_split = value_rowsplit(0);
+  for (int i = 1; i < tensor_length; ++i) {
+    const int split = value_rowsplit(i);
+    max_width = std::max(max_width, split - prev_split);
+    prev_split = split;
+  }
+  return max_width;
+}
+
+int GetMaxWidth(const ConversionAttributes& conversion_attributes,
+                TfLiteContext* context, TfLiteNode* node, int dimension) {
+  const TfLiteTensor* tensor = GetRowPartitionTensor(
+      conversion_attributes, context, node, dimension - 1);
+  switch (conversion_attributes.GetRowPartitionTypeByDimension(dimension - 1)) {
+    case tensorflow::RowPartitionType::VALUE_ROWIDS:
+      return GetMaxWidthValueRowID(tensor);
+    case tensorflow::RowPartitionType::ROW_SPLITS:
+      return GetMaxWidthRowSplit(tensor);
+    default:
+      context->ReportError(context, "Cannot handle partition type");
+      return -1;
+  }
+}
+
+RuntimeShape CombineRaggedTensorToTensorShapes(
+    int ragged_rank, const RuntimeShape& output_shape,
+    const RuntimeShape& value_shape) {
+  // TODO(mgubin): No checks, see
+  // third_party/tensorflow/core/ops/ragged_to_dense_util.cc
+  RuntimeShape result(output_shape);
+  if (output_shape.DimensionsCount() == 0) {
+    const int output_shape_rank = ragged_rank + value_shape.DimensionsCount();
+    result.Resize(output_shape_rank);
+    for (int i = 0; i < output_shape_rank; ++i) {
+      result.SetDim(i, -1);
+    }
+  }
+  const int need_to_set =
+      output_shape.DimensionsCount() - value_shape.DimensionsCount();
+  for (int i = 1; i < value_shape.DimensionsCount(); ++i) {
+    result.SetDim(need_to_set + i, value_shape.Dims(i));
+  }
+  return result;
+}
+
+RuntimeShape CalculateOutputSize(
+    const ConversionAttributes& conversion_attributes, TfLiteContext* context,
+    TfLiteNode* node, int first_dimension, int ragged_rank,
+    const TfLiteTensor& values, const TfLiteTensor& default_value,
+    const TfLiteTensor& output_shape) {
+  RuntimeShape values_shape(values.dims->size, values.dims->data);
+  RuntimeShape default_value_shape(default_value.dims->size,
+                                   default_value.dims->data);
+
+  if (!ValidateDefaultValueShape(context, default_value_shape, values_shape)) {
+    return {};
+  }
+  RuntimeShape output_shape_shape = TensorShapeFromTensor(output_shape);
+
+  RuntimeShape result_shape = CombineRaggedTensorToTensorShapes(
+      ragged_rank, output_shape_shape, values_shape);
+  if (result_shape.Dims(0) < 0) {
+    result_shape.SetDim(0, first_dimension);
+  }
+  for (int i = 1; i <= ragged_rank; ++i) {
+    if (result_shape.Dims(i) < 0) {
+      result_shape.SetDim(i,
+                          GetMaxWidth(conversion_attributes, context, node, i));
+    }
+  }
+  return result_shape;
+}
+
+TfLiteIntArray* IntArrayFromShape(const RuntimeShape& shape) {
+  TfLiteIntArray* result = TfLiteIntArrayCreate(shape.DimensionsCount());
+  for (int i = 0; i < shape.DimensionsCount(); ++i) {
+    result->data[i] = shape.Dims(i);
+  }
+  return result;
+}
+
+/**
+ * The output_index represents the index in the output tensor
+ * where the first element of a particular dimension would be written.
+ * If it is -1, it indicates that the index is out of scope.
+ * Example, given first_dimension = 10, first_dimension_output = 6,
+ * and output_index_multiplier = 100:
+ * result = [0 100 200 300 400 500 -1 -1 -1 -1]
+ * If first_dimension_output = 11 instead, then:
+ * result = [0 100 200 300 400 500 600 700 800 900]
+ */
+void CalculateFirstParentOutputIndex(int first_dimension,
+                                     int output_index_multiplier,
+                                     int first_dimension_output,
+                                     std::vector<int>* result) {
+  const int min_dimension = std::min(first_dimension, first_dimension_output);
+  result->reserve(first_dimension);
+  int current_output_index = 0;
+  for (int i = 0; i < min_dimension;
+       ++i, current_output_index += output_index_multiplier) {
+    result->push_back(current_output_index);
+  }
+  for (int i = min_dimension; i < first_dimension; ++i) {
+    result->push_back(-1);
+  }
+}
+// Calculate the output index of the first element of a list.
+// The parent_output_index is the same computation for the previous list.
+// -1 indicates an element or list that is out of range.
+// The output_index_multiplier is the number of output indices one moves
+// forward for each column.
+// E.g., given:
+// value_rowids:[0 1 2 2 2 3 5 5 6]
+// parent_output_index:[1000 1100 2000 2100 -1 3000 4000]
+// output_index_multiplier: 10
+// output_size: 2
+// You get:
+// result = [1000 1100 2000 2010 -1 2100 -1 -1 3000]
+// result[0] = parent_output_index[value_rowids[0]]
+// result[1] = parent_output_index[value_rowids[1]]
+// result[2] = parent_output_index[value_rowids[2]]
+// result[3] = parent_output_index[value_rowids[2] + 10]
+// result[4] = -1 because it is the third element the size is 2.
+// result[5] = parent_output_index[value_rowids[3]]
+// result[6] = -1 because parent_output_index[value_rowids[6]] == -1
+// result[7] = -1 because parent_output_index[value_rowids[6]] == -1
+// result[8] = parent_output_index[value_rowids[7]]
+void CalculateOutputIndexValueRowID(const TfLiteTensor& value_rowids,
+                                    const std::vector<int>& parent_output_index,
+                                    int output_index_multiplier,
+                                    int output_size, std::vector<int>* result) {
+  const RuntimeShape tensor_shape(value_rowids.dims->size,
+                                  value_rowids.dims->data);
+  const int index_size = tensor_shape.FlatSize();
+  result->reserve(index_size);
+  if (index_size == 0) {
+    return;
+  }
+
+  auto value_rowids_val = [value_rowids](int index) {
+    switch (value_rowids.type) {
+      case kTfLiteInt32:
+        return static_cast<int>(value_rowids.data.i32[index]);
+      case kTfLiteInt64:
+        return static_cast<int>(value_rowids.data.i64[index]);
+      default:
+        // TODO(mgubin): Add error checks.
+        return 0;
+    }
+  };
+  int current_output_column = 0;
+  int current_value_rowid = value_rowids_val(0);
+  // DCHECK_LT(current_value_rowid, parent_output_index.size());
+  int current_output_index = parent_output_index[current_value_rowid];
+  result->push_back(current_output_index);
+  for (int i = 1; i < index_size; ++i) {
+    int next_value_rowid = value_rowids_val(i);
+    if (next_value_rowid == current_value_rowid) {
+      if (current_output_index >= 0) {
+        ++current_output_column;
+        if (current_output_column < output_size) {
+          current_output_index += output_index_multiplier;
+        } else {
+          current_output_index = -1;
+        }
+      }
+    } else {
+      current_output_column = 0;
+      current_value_rowid = next_value_rowid;
+      // DCHECK_LT(next_value_rowid, parent_output_index.size());
+      current_output_index = parent_output_index[next_value_rowid];
+    }
+    result->push_back(current_output_index);
+  }
+  // DCHECK_EQ(result->size(), value_rowids.size());
+}
+
+void CalculateOutputIndexRowSplit(const TfLiteTensor& row_split,
+                                  const std::vector<int>& parent_output_index,
+                                  int output_index_multiplier, int output_size,
+                                  std::vector<int>* result) {
+  const RuntimeShape row_split_shape(row_split.dims->size,
+                                     row_split.dims->data);
+  const int row_split_size = row_split_shape.FlatSize();
+  auto row_split_val = [row_split](int index) {
+    switch (row_split.type) {
+      case kTfLiteInt32:
+        return static_cast<int>(row_split.data.i32[index]);
+      case kTfLiteInt64:
+        return static_cast<int>(row_split.data.i64[index]);
+      default:
+        // TODO(mgubin): Add error checks.
+        return 0;
+    }
+  };
+  if (row_split_size > 0) {
+    result->reserve(row_split_val(row_split_size - 1));
+  }
+  for (int i = 0; i < row_split_size - 1; ++i) {
+    const int row_length = row_split_val(i + 1) - row_split_val(i);
+    int real_length = std::min(output_size, row_length);
+    int parent_output_index_current = parent_output_index[i];
+
+    if (parent_output_index_current == -1) {
+      real_length = 0;
+    }
+    for (int j = 0; j < real_length; ++j) {
+      result->push_back(parent_output_index_current);
+      parent_output_index_current += output_index_multiplier;
+    }
+    for (int j = 0; j < row_length - real_length; ++j) {
+      result->push_back(-1);
+    }
+  }
+  // if (row_split_size > 0) {
+  //  DCHECK_EQ(result->size(), row_split(row_split_size - 1));
+  //}
+}
+
+TfLiteStatus CalculateOutputIndex(
+    const ConversionAttributes& conversion_attributes, TfLiteContext* context,
+    TfLiteNode* node, int dimension,
+    const std::vector<int>& parent_output_index, int output_index_multiplier,
+    int output_size, std::vector<int>* result) {
+  const TfLiteTensor* row_partition_tensor =
+      GetRowPartitionTensor(conversion_attributes, context, node, dimension);
+  auto partition_type =
+      conversion_attributes.GetRowPartitionTypeByDimension(dimension);
+  switch (partition_type) {
+    case tensorflow::RowPartitionType::VALUE_ROWIDS:
+      CalculateOutputIndexValueRowID(*row_partition_tensor, parent_output_index,
+                                     output_index_multiplier, output_size,
+                                     result);
+      return kTfLiteOk;
+    case tensorflow::RowPartitionType::ROW_SPLITS:
+      CalculateOutputIndexRowSplit(*row_partition_tensor, parent_output_index,
+                                   output_index_multiplier, output_size,
+                                   result);
+      return kTfLiteOk;
+    default:
+      context->ReportError(context, "Unsupported partition type");
+      return kTfLiteError;
+  }
+}
+
+template <typename VALUE_TYPE>
+void SetOutputT(TfLiteContext* context, int ragged_rank,
+                const std::vector<int>& output_index,
+                const TfLiteTensor& values_tensor,
+                const TfLiteTensor& default_value_tensor,
+                TfLiteTensor* output_tensor) {
+  const VALUE_TYPE* values_base = GetTensorData<VALUE_TYPE>(&values_tensor);
+  VALUE_TYPE* output_base = GetTensorData<VALUE_TYPE>(output_tensor);
+  const VALUE_TYPE* default_value =
+      GetTensorData<VALUE_TYPE>(&default_value_tensor);
+
+  RuntimeShape output_shape = GetTensorShape(output_tensor);
+  RuntimeShape element_shape =
+      RuntimeShape(output_shape.DimensionsCount() - ragged_rank - 1,
+                   output_shape.DimsData() + ragged_rank + 1);
+
+  // element_shape.RemoveDimRange(0, ragged_rank + 1);
+  const int value_element_size = element_shape.FlatSize();
+  size_t output_index_size = output_index.size();
+
+  // Loop through the output_index vector, finding contiguous regions that
+  // should be copied.  Once we find the end of a contiguous region, copy it
+  // and add any necessary padding (with default_value).
+  int src_start = 0;  // Start of contiguous region (in values)
+  int dst_start = 0;  // Destination for contiguous region (in output)
+  int dst_end = 0;    // Destination for contiguous region (in output)
+  for (int src_i = 0; src_i <= output_index_size; ++src_i) {
+    // dst_i is the destination where the value at src_i should be copied.
+    int dst_i = src_i < output_index_size ? output_index[src_i] : -1;
+
+    // If we're still in a contiguous region, then update dst_end go to the
+    // next src_i.
+    if (dst_i == dst_end) {
+      ++dst_end;
+      continue;
+    }
+
+    // We found the end of contiguous region.  This can be because we found
+    // a gap (dst_i > dst_end), or a source value that shouldn't be copied
+    // because it's out-of-bounds (dst_i == -1), or the end of the tensor
+    // (dst_i = -1).
+    if (dst_start < dst_end) {
+      // Copy the contiguous region.
+      const VALUE_TYPE* src = values_base + src_start * value_element_size;
+      VALUE_TYPE* dst = output_base + dst_start * value_element_size;
+      int nvals = (dst_end - dst_start) * value_element_size;
+      std::copy(src, src + nvals, dst);
+      // copy_array<VALUE_TYPE, int>(dst, src, nvals);
+    }
+
+    // Add any necessary padding (w/ default_value).
+    if (src_i >= output_index_size) {
+      // We reached the end of values: pad to the end of output.
+      const int output_size = output_shape.FlatSize();
+      dst_i = output_size / value_element_size;
+    }
+    if (dst_i > dst_end) {
+      std::fill(output_base + dst_end * value_element_size,
+                output_base + dst_i * value_element_size, *default_value);
+      dst_end = dst_i;
+    }
+
+    // Update indices.
+    if (dst_i < 0) {
+      // src_i should be skipped -- leave it out of the contiguous region.
+      src_start = src_i + 1;
+      dst_start = dst_end;
+    } else {
+      // src_i should be copied -- include it in the contiguous region.
+      src_start = src_i;
+      dst_start = dst_end;
+      dst_end = dst_start + 1;
+    }
+  }
+}
+
+void SetOutput(TfLiteContext* context, int ragged_rank,
+               const std::vector<int>& output_index,
+               const TfLiteTensor& values_tensor,
+               const TfLiteTensor& default_value_tensor,
+               TfLiteTensor* output_tensor) {
+  switch (output_tensor->type) {
+    case kTfLiteInt32:
+      SetOutputT<int32_t>(context, ragged_rank, output_index, values_tensor,
+                          default_value_tensor, output_tensor);
+      break;
+    case kTfLiteInt64:
+      SetOutputT<int64_t>(context, ragged_rank, output_index, values_tensor,
+                          default_value_tensor, output_tensor);
+      break;
+    case kTfLiteFloat32:
+      SetOutputT<float>(context, ragged_rank, output_index, values_tensor,
+                        default_value_tensor, output_tensor);
+      break;
+    default:
+      context->ReportError(context, "Not supported values type");
+  }
+}
+
+}  // namespace
+
+void* Initialize(TfLiteContext* context, const char* buffer, size_t length) {
+  auto attributes = std::make_unique<ConversionAttributes>();
+
+  const uint8_t* buffer_t = reinterpret_cast<const uint8_t*>(buffer);
+
+  const flexbuffers::Map& m = flexbuffers::GetRoot(buffer_t, length).AsMap();
+  // TODO (mgubin): Converting flat buffer to a vector of strings looks not very
+  // effective but simple. A cleaner way is needed.
+  const flexbuffers::TypedVector row_partition_types_attr =
+      m[kRowPartitionTypesAttr].AsTypedVector();
+  std::vector<std::string> row_partition_types_attr_strings;
+  row_partition_types_attr_strings.reserve(row_partition_types_attr.size());
+  for (int i = 0; i < row_partition_types_attr.size(); ++i) {
+    row_partition_types_attr_strings.emplace_back(
+        row_partition_types_attr[i].AsString().str());
+  }
+  attributes->partition_types =
+      tensorflow::GetRowPartitionTypesHelper(row_partition_types_attr_strings);
+  if (attributes->partition_types.size() !=
+      row_partition_types_attr_strings.size()) {
+    context->ReportError(context, "Can't parse partition type attribute");
+    return nullptr;
+  }
+  attributes->ragged_rank =
+      tensorflow::GetRaggedRank(attributes->partition_types);
+  return attributes.release();
+}
+void Free(TfLiteContext* /*context*/, void* buffer) {
+  ConversionAttributes* attributes =
+      reinterpret_cast<ConversionAttributes*>(buffer);
+  delete attributes;
+}
+
+TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
+  const ConversionAttributes* attributes =
+      reinterpret_cast<ConversionAttributes*>(node->user_data);
+  if (attributes == nullptr) {
+    // Parsing attributes failed, can't prepare.
+    context->ReportError(context, "Attributes are not initialized");
+    return kTfLiteError;
+  }
+  // The output tensor need to be set to dynamic because it can have different
+  // size.
+  TfLiteTensor& output_tensor =
+      context->tensors[node->outputs->data[kOutputTensor]];
+  SetTensorToDynamic(&output_tensor);
+
+  // Check that input shape tensor is int32 or int64
+  TfLiteTensor& input_shape = context->tensors[node->inputs->data[kShapeInput]];
+  if (input_shape.type != kTfLiteInt32 && input_shape.type != kTfLiteInt64) {
+    context->ReportError(context,
+                         "Input form tensor could be only int32 or int64");
+    return kTfLiteError;
+  }
+  return kTfLiteOk;
+}
+
+TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
+  const ConversionAttributes* attributes =
+      reinterpret_cast<ConversionAttributes*>(node->user_data);
+  TfLiteTensor& input_shape = context->tensors[node->inputs->data[kShapeInput]];
+  TfLiteTensor& input_values =
+      context->tensors[node->inputs->data[kValuesInput]];
+  TfLiteTensor& default_value =
+      context->tensors[node->inputs->data[kDefaultValueInput]];
+  // TODO (mgubin): Only scallar default value is supported.
+  if (RuntimeShape(default_value.dims->size, default_value.dims->data)
+          .FlatSize() != 1) {
+    context->ReportError(context, "Only scallar default value is supported");
+    return kTfLiteError;
+  }
+  TfLiteTensor& first_partition_input =
+      context->tensors[node->inputs->data[kFirstPartitionInputIndex]];
+
+  // Calculate dimensions.
+  const int first_dimension =
+      GetFirstDimensionSize(context, first_partition_input, attributes);
+  if (first_dimension < 0) {
+    return kTfLiteError;
+  }
+  RuntimeShape output_shape = CalculateOutputSize(
+      *attributes, context, node, first_dimension, attributes->ragged_rank,
+      input_values, default_value, input_shape);
+  if (output_shape.DimensionsCount() == 0) {
+    return kTfLiteError;
+  }
+
+  std::vector<int> multiplier;
+  multiplier.resize(attributes->ragged_rank + 1);
+  multiplier.back() = 1;
+  for (int i = multiplier.size() - 2; i >= 0; --i) {
+    multiplier[i] = multiplier[i + 1] * output_shape.Dims(i + 1);
+  }
+
+  // Allocate output tensor.
+  TfLiteTensor& output_tensor =
+      context->tensors[node->outputs->data[kOutputTensor]];
+
+  TF_LITE_ENSURE_OK(context,
+                    context->ResizeTensor(context, &output_tensor,
+                                          IntArrayFromShape(output_shape)));
+
+  // Copy data.
+  const int full_size = multiplier.front() * output_shape.Dims(0);
+  if (full_size > 0) {
+    std::vector<int> output_index, new_output_index;
+    int nvals = input_values.dims->data[0];
+    output_index.reserve(nvals);
+    new_output_index.reserve(nvals);
+
+    CalculateFirstParentOutputIndex(first_dimension, multiplier[0],
+                                    output_shape.Dims(0), &output_index);
+    for (int i = 1; i <= attributes->ragged_rank; ++i) {
+      TF_LITE_ENSURE_OK(
+          context, CalculateOutputIndex(
+                       *attributes, context, node, i - 1, output_index,
+                       multiplier[i], output_shape.Dims(i), &new_output_index));
+      output_index.swap(new_output_index);
+      new_output_index.clear();
+    }
+
+    SetOutput(context, attributes->ragged_rank, output_index, input_values,
+              default_value, &output_tensor);
+  }
+  return kTfLiteOk;
+}
+
+}  // namespace ragged_tensor_to_tensor
+}  // namespace ragged
+
+TfLiteRegistration* Register_RAGGED_TENSOR_TO_TENSOR() {
+  static TfLiteRegistration r = {ragged::ragged_tensor_to_tensor::Initialize,
+                                 ragged::ragged_tensor_to_tensor::Free,
+                                 ragged::ragged_tensor_to_tensor::Prepare,
+                                 ragged::ragged_tensor_to_tensor::Eval};
+  return &r;
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_tensor_to_tensor_tflite_test.cc b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_tensor_to_tensor_tflite_test.cc
new file mode 100644
index 0000000..b1cde57
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/ragged/ragged_tensor_to_tensor_tflite_test.cc
@@ -0,0 +1,283 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <initializer_list>
+#include <string>
+#include <vector>
+
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include "flatbuffers/flexbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/interpreter.h"
+#include "tensorflow/lite/kernels/internal/tensor.h"
+#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
+#include "tensorflow/lite/kernels/test_util.h"
+#include "tensorflow/lite/schema/schema_generated.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+TfLiteRegistration* Register_RAGGED_TENSOR_TO_TENSOR();
+}  // namespace custom
+}  // namespace ops
+
+namespace {
+
+class RaggedTensorToTensorOpModel : public SingleOpModel {
+ public:
+  RaggedTensorToTensorOpModel(int output_shape_dims,
+                              std::initializer_list<int> values_shape,
+                              std::initializer_list<std::initializer_list<int>>
+                                  partition_tensors_shapes,
+                              std::vector<std::string> partition_types,
+                              TensorType value_type = TensorType_FLOAT32,
+                              TensorType index_type = TensorType_INT32) {
+    // A structure to collect shapes for the input.
+    std::vector<std::vector<int>> shapes;
+    input_shape_ = AddInput(index_type);
+    shapes.push_back({output_shape_dims});
+    input_values_ = AddInput(value_type);
+    shapes.emplace_back(values_shape);
+    input_default_values_ = AddInput(value_type);
+    shapes.push_back({1});
+    for (const auto& p : partition_tensors_shapes) {
+      partition_tensors_.push_back(AddInput(TensorType_INT32));
+      shapes.emplace_back(p);
+    }
+    output_ = AddOutput(value_type);
+
+    flexbuffers::Builder fbb;
+    size_t start = fbb.StartMap();
+    {
+      size_t start = fbb.StartVector("row_partition_types");
+      for (const auto& s : partition_types) {
+        fbb.String(s);
+      }
+      fbb.EndVector(start, /*typed=*/true, /*fixed=*/false);
+    }
+    fbb.Int("num_row_partition_tensors", partition_types.size());
+    fbb.EndMap(start);
+    fbb.Finish();
+    SetCustomOp("RaggedTensorToTensor", fbb.GetBuffer(),
+                ops::custom::Register_RAGGED_TENSOR_TO_TENSOR);
+    BuildInterpreter(shapes);
+  }
+
+  std::vector<int> GetOutputShape() { return GetTensorShape(output_); }
+
+  std::vector<float> GetOutputFloat() { return ExtractVector<float>(output_); }
+  std::vector<int32> GetOutputInt() { return ExtractVector<int32>(output_); }
+
+  void InvokeFloat(const std::vector<int>& shape,
+                   const std::vector<float>& values, float default_value,
+                   const std::vector<std::vector<int>>& partition_values) {
+    PopulateTensor(input_shape_, shape);
+    PopulateTensor(input_values_, values);
+    PopulateTensor(input_default_values_, {default_value});
+    for (int i = 0; i < partition_values.size(); ++i) {
+      PopulateTensor(partition_tensors_[i], partition_values[i]);
+    }
+    SingleOpModel::Invoke();
+  }
+  void InvokeInt(const std::vector<int>& shape,
+                 const std::vector<int32>& values, int32 default_value,
+                 const std::vector<std::vector<int>>& partition_values) {
+    PopulateTensor(input_shape_, shape);
+    PopulateTensor(input_values_, values);
+    PopulateTensor(input_default_values_, {default_value});
+    for (int i = 0; i < partition_values.size(); ++i) {
+      PopulateTensor(partition_tensors_[i], partition_values[i]);
+    }
+    SingleOpModel::Invoke();
+  }
+
+ private:
+  int input_shape_;
+  int input_values_;
+  int input_default_values_;
+  std::vector<int> partition_tensors_;
+  int output_;
+};
+
+TEST(RaggedTensorToTensorTest, RaggedTensorToTensor) {
+  // indices = [2, 1, 0, 3]
+  // params = [[.1, .2, .3], [], [.4, .5, .6, .7], [.8, .9]]
+  // params.shape = [4, None]
+  RaggedTensorToTensorOpModel model(
+      2,           // output_shape_dims
+      {9},         // values_shape
+      {{1}, {9}},  // partition_tensors_shapes
+      std::vector<std::string>({"FIRST_DIM_SIZE", "VALUE_ROWIDS"}));
+  model.InvokeFloat({4, 4},                                // shape
+                    {.1, .2, .3, .4, .5, .6, .7, .8, .9},  // values
+                    1.5,                                   // default_value
+                    std::vector<std::vector<int>>(
+                        {std::vector<int>({4}),
+                         std::vector<int>({0, 0, 0, 2, 2, 2, 2, 3, 3})}));
+  EXPECT_THAT(model.GetOutputShape(), testing::ElementsAreArray({4, 4}));
+  EXPECT_THAT(model.GetOutputFloat(),
+              testing::ElementsAreArray({.1, .2, .3, 1.5, 1.5, 1.5, 1.5, 1.5,
+                                         .4, .5, .6, .7, .8, .9, 1.5, 1.5}));
+}
+
+TEST(RaggedTensorToTensorTest, RaggedTensorToTensorRowSplits) {
+  // indices = [2, 1, 0, 3]
+  // params = [[.1, .2, .3], [], [.4, .5, .6, .7], [.8, .9]]
+  RaggedTensorToTensorOpModel model(2,      // output_shape_dims
+                                    {9},    // values_shape
+                                    {{5}},  // partition_tensors_shapes
+                                    std::vector<std::string>({"ROW_SPLITS"}));
+  model.InvokeFloat(
+      {4, 4},                                // shape
+      {.1, .2, .3, .4, .5, .6, .7, .8, .9},  // values
+      1.5,                                   // default_value
+      std::vector<std::vector<int>>({std::vector<int>({0, 3, 3, 7, 9})}));
+  EXPECT_THAT(model.GetOutputShape(), testing::ElementsAreArray({4, 4}));
+  EXPECT_THAT(model.GetOutputFloat(),
+              testing::ElementsAreArray({.1, .2, .3, 1.5, 1.5, 1.5, 1.5, 1.5,
+                                         .4, .5, .6, .7, .8, .9, 1.5, 1.5}));
+}
+
+TEST(RaggedTensorToTensorTest, RaggedTensorToTensor_3DParams) {
+  // params = [
+  //           [[]],
+  //           [[.1, .2], [.3]],
+  //           [],
+  //           [[.4, .5], [.6, .7, .8]],
+  //           [[.9]]
+  //          ]
+  RaggedTensorToTensorOpModel model(
+      3,                // output_shape_dims
+      {9},              // values_shape
+      {{1}, {6}, {9}},  // partition_tensors_shapes
+      std::vector<std::string>(
+          {"FIRST_DIM_SIZE", "VALUE_ROWIDS", "VALUE_ROWIDS"}));
+  model.InvokeFloat(
+      {5, 2, 3},                             // shape
+      {.1, .2, .3, .4, .5, .6, .7, .8, .9},  // values
+      1.5,                                   // default_value
+      std::vector<std::vector<int>>(
+          {std::vector<int>({5}), std::vector<int>({0, 1, 1, 3, 3, 4}),
+           std::vector<int>({1, 1, 2, 3, 3, 4, 4, 4, 5})}));
+
+  EXPECT_THAT(model.GetOutputShape(), testing::ElementsAreArray({5, 2, 3}));
+  EXPECT_THAT(model.GetOutputFloat(),
+              testing::ElementsAreArray({1.5, 1.5, 1.5, 1.5, 1.5, 1.5, .1,  .2,
+                                         1.5, .3,  1.5, 1.5, 1.5, 1.5, 1.5, 1.5,
+                                         1.5, 1.5, .4,  .5,  1.5, .6,  .7,  .8,
+                                         .9,  1.5, 1.5, 1.5, 1.5, 1.5}));
+}
+
+TEST(RaggedTensorToTensorOpTest, RaggedTensorToTensor_3DParamsRowSplits) {
+  // params = [
+  //           [[]],
+  //           [[.1, .2], [.3]],
+  //           [],
+  //           [[.4, .5], [.6, .7, .8]],
+  //           [[.9]]
+  //          ]
+  RaggedTensorToTensorOpModel model(
+      3,           // output_shape_dims
+      {9},         // values_shape
+      {{6}, {7}},  // partition_tensors_shapes
+      std::vector<std::string>({"ROW_SPLITS", "ROW_SPLITS"}));
+  model.InvokeFloat(
+      {5, 2, 3},                             // shape
+      {.1, .2, .3, .4, .5, .6, .7, .8, .9},  // values
+      1.5,                                   // default_value
+      std::vector<std::vector<int>>({std::vector<int>({0, 1, 3, 3, 5, 6}),
+                                     std::vector<int>({0, 0, 2, 3, 5, 8, 9})}));
+  EXPECT_THAT(model.GetOutputShape(), testing::ElementsAreArray({5, 2, 3}));
+  EXPECT_THAT(model.GetOutputFloat(),
+              testing::ElementsAreArray({1.5, 1.5, 1.5, 1.5, 1.5, 1.5, .1,  .2,
+                                         1.5, .3,  1.5, 1.5, 1.5, 1.5, 1.5, 1.5,
+                                         1.5, 1.5, .4,  .5,  1.5, .6,  .7,  .8,
+                                         .9,  1.5, 1.5, 1.5, 1.5, 1.5}));
+}
+
+TEST(RaggedTensorToTensorTest, RaggedTensorToTensor_3DParamsRowSplits2) {
+  // params = [
+  //           [[0, 1, 2], []],
+  //           [],
+  //           [[3]]
+  //          ]
+
+  RaggedTensorToTensorOpModel model(
+      3,           // output_shape_dims
+      {4},         // values_shape
+      {{4}, {4}},  // partition_tensors_shapes
+      std::vector<std::string>({"ROW_SPLITS", "ROW_SPLITS"}), TensorType_INT32);
+  model.InvokeInt(
+      {3, 2, 3},     // shape
+      {0, 1, 2, 3},  // values
+      5,             // default_value
+      std::vector<std::vector<int>>(
+          {std::vector<int>({0, 2, 2, 3}), std::vector<int>({0, 3, 3, 4})}));
+
+  EXPECT_THAT(model.GetOutputShape(), testing::ElementsAreArray({3, 2, 3}));
+
+  EXPECT_THAT(model.GetOutputInt(),
+              testing::ElementsAreArray(
+                  {0, 1, 2, 5, 5, 5, 5, 5, 5, 5, 5, 5, 3, 5, 5, 5, 5, 5}));
+}
+
+TEST(RaggedTensorToTensorTest, RaggedTensorToTensorContractExpanded) {
+  // params = [[.1, .2, .3], [], [.4, .5, .6, .7], [.8, .9]]
+  RaggedTensorToTensorOpModel model(
+      2,           // output_shape_dims
+      {9},         // values_shape
+      {{1}, {9}},  // partition_tensors_shapes
+      std::vector<std::string>({"FIRST_DIM_SIZE", "VALUE_ROWIDS"}));
+  model.InvokeFloat({3, 5},                                // shape
+                    {.1, .2, .3, .4, .5, .6, .7, .8, .9},  // values
+                    1.5,                                   // default_value
+                    std::vector<std::vector<int>>(
+                        {std::vector<int>({4}),
+                         std::vector<int>({0, 0, 0, 2, 2, 2, 2, 3, 3})}));
+  EXPECT_THAT(model.GetOutputShape(), testing::ElementsAreArray({3, 5}));
+
+  EXPECT_THAT(model.GetOutputFloat(),
+              testing::ElementsAreArray({.1, .2, .3, 1.5, 1.5,     //
+                                         1.5, 1.5, 1.5, 1.5, 1.5,  //
+                                         .4, .5, .6, .7, 1.5}));
+}
+
+// Adds a dense dimension.
+TEST(RaggedTensorToTensorTest, RaggedTensorToTensorContractExpandedDense) {
+  // params = [[.1, .2, .3], [], [.4, .5, .6, .7], [.8, .9]]
+  RaggedTensorToTensorOpModel model(
+      3,           // output_shape_dims
+      {9, 2},      // values_shape
+      {{1}, {9}},  // partition_tensors_shapes
+      std::vector<std::string>({"FIRST_DIM_SIZE", "VALUE_ROWIDS"}));
+
+  model.InvokeFloat({3, 5, 2},  // shape
+                    {.1, 1.1, .2, 1.2, .3, 1.3, .4, 1.4, .5, 1.5, .6, 1.6, .7,
+                     1.7, .8, 1.8, .9, 1.9},  // values
+                    1.5,                      // default_value
+                    std::vector<std::vector<int>>(
+                        {std::vector<int>({4}),
+                         std::vector<int>({0, 0, 0, 2, 2, 2, 2, 3, 3})}));
+
+  EXPECT_THAT(model.GetOutputShape(), testing::ElementsAreArray({3, 5, 2}));
+  EXPECT_THAT(model.GetOutputFloat(),
+              testing::ElementsAreArray(
+                  {.1,  1.1, .2,  1.2, .3,  1.3, 1.5, 1.5, 1.5, 1.5,  //
+                   1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5, 1.5,  //
+                   .4,  1.4, .5,  1.5, .6,  1.6, .7,  1.7, 1.5, 1.5}));
+}
+}  // namespace
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/BUILD b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/BUILD
new file mode 100644
index 0000000..e8df50f
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/BUILD
@@ -0,0 +1,389 @@
+# Memorymappable, WASM compilable, implementation of the encoder.
+#
+
+load("@flatbuffers//:build_defs.bzl", "flatbuffer_cc_library")
+load("@org_tensorflow//tensorflow:tensorflow.bzl", "tf_gen_op_wrapper_py")
+load(":native.bzl", "micore_tf_copts", "micore_tf_deps")
+
+package(
+    default_visibility = [
+        "//tensorflow_lite_support:users",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+filegroup(
+    name = "testdata",
+    srcs = glob([
+        "testdata/**",
+    ]),
+)
+
+filegroup(
+    name = "config_fbs",
+    srcs = ["config.fbs"],
+)
+
+flatbuffer_cc_library(
+    name = "config",
+    srcs = [
+        "config.fbs",
+    ],
+)
+
+flatbuffer_cc_library(
+    name = "encoder_config",
+    srcs = [
+        "encoder_config.fbs",
+    ],
+    includes = [":config_fbs"],
+)
+
+flatbuffer_cc_library(
+    name = "decoder_config",
+    srcs = [
+        "decoder_config.fbs",
+    ],
+    includes = [":config_fbs"],
+)
+
+cc_library(
+    name = "utils",
+    srcs = [
+    ],
+    hdrs = [
+        "utils.h",
+    ],
+)
+
+cc_library(
+    name = "double_array_trie",
+    srcs = [
+    ],
+    hdrs = [
+        "double_array_trie.h",
+    ],
+    deps = [
+        ":config",
+        ":utils",
+    ],
+)
+
+cc_library(
+    name = "double_array_trie_builder",
+    srcs = [
+        "double_array_trie_builder.cc",
+    ],
+    hdrs = [
+        "double_array_trie_builder.h",
+    ],
+    deps = [
+        ":config",
+        ":utils",
+        "@darts_clone",
+    ],
+)
+
+cc_test(
+    name = "double_array_trie_test",
+    srcs = [
+        "double_array_trie_test.cc",
+    ],
+    deps = [
+        ":double_array_trie",
+        ":double_array_trie_builder",
+        ":encoder_config",
+        "@com_google_googletest//:gtest_main",
+    ],
+)
+
+cc_library(
+    name = "sentencepiece_constants",
+    srcs = [],
+    hdrs = ["sentencepiece_constants.h"],
+)
+
+cc_library(
+    name = "model_converter",
+    srcs = [
+        "model_converter.cc",
+    ],
+    hdrs = [
+        "model_converter.h",
+    ],
+    deps = [
+        ":config",
+        ":decoder_config",
+        ":double_array_trie_builder",
+        ":encoder_config",
+        ":sentencepiece_constants",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_sentencepiece//src:sentencepiece_model_cc_proto",
+    ],
+)
+
+cc_library(
+    name = "optimized_encoder",
+    srcs = [
+        "optimized_encoder.cc",
+    ],
+    hdrs = [
+        "optimized_encoder.h",
+    ],
+    deps = [
+        ":config",
+        ":double_array_trie",
+        ":encoder_config",
+    ],
+)
+
+cc_library(
+    name = "optimized_decoder",
+    srcs = [
+        "optimized_decoder.cc",
+    ],
+    hdrs = [
+        "optimized_decoder.h",
+    ],
+    deps = [
+        "config",
+        ":decoder_config",
+        ":double_array_trie",
+    ],
+)
+
+cc_library(
+    name = "sentencepiece_tokenizer_h",
+    hdrs = [
+        "sentencepiece_tokenizer.h",
+    ],
+)
+
+cc_library(
+    name = "sentencepiece_detokenizer_h",
+    hdrs = [
+        "sentencepiece_detokenizer.h",
+    ],
+)
+
+cc_library(
+    name = "sentencepiece_tokenizer_op",
+    srcs = ["sentencepiece_tokenizer_op.cc"],
+    copts = micore_tf_copts(),
+    visibility = [
+        "//visibility:public",
+    ],
+    deps = [
+        ":sentencepiece_tokenizer_h",
+        ":optimized_encoder",
+    ] + micore_tf_deps(),
+    alwayslink = 1,
+)
+
+cc_binary(
+    name = "sentencepiece_tokenizer_op.so",
+    srcs = [
+        "sentencepiece_tokenizer_op.cc",
+    ],
+    copts = micore_tf_copts(),
+    linkshared = 1,
+    deps = [
+        ":sentencepiece_tokenizer_h",
+        ":optimized_encoder",
+    ] + micore_tf_deps(),
+)
+
+cc_library(
+    name = "sentencepiece_detokenizer_op",
+    srcs = ["sentencepiece_detokenizer_op.cc"],
+    copts = micore_tf_copts(),
+    visibility = [
+        "//visibility:public",
+    ],
+    deps = [
+        ":sentencepiece_detokenizer_h",
+        ":optimized_decoder",
+    ] + micore_tf_deps(),
+    alwayslink = 1,
+)
+
+cc_binary(
+    name = "sentencepiece_detokenizer_op.so",
+    srcs = [
+        "sentencepiece_detokenizer_op.cc",
+    ],
+    copts = micore_tf_copts(),
+    linkshared = 1,
+    deps = [
+        ":sentencepiece_detokenizer_h",
+        ":optimized_decoder",
+    ] + micore_tf_deps(),
+)
+
+cc_library(
+    name = "sentencepiece_tokenizer_tflite",
+    srcs = ["sentencepiece_tokenizer_tflite.cc"],
+    visibility = [
+        "//visibility:public",
+    ],
+    deps =
+        [
+            ":optimized_encoder",
+            ":sentencepiece_tokenizer_h",
+            "@flatbuffers",
+            "@org_tensorflow//tensorflow/lite:framework",
+            "@org_tensorflow//tensorflow/lite:string_util",
+            "@org_tensorflow//tensorflow/lite/c:common",
+            "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+            "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+        ],
+)
+
+cc_library(
+    name = "sentencepiece_detokenizer_tflite",
+    srcs = ["sentencepiece_detokenizer_tflite.cc"],
+    visibility = [
+        "//visibility:public",
+    ],
+    deps =
+        [
+            ":optimized_decoder",
+            ":sentencepiece_detokenizer_h",
+            "@flatbuffers",
+            "@org_tensorflow//tensorflow/lite:framework",
+            "@org_tensorflow//tensorflow/lite:string_util",
+            "@org_tensorflow//tensorflow/lite/c:common",
+            "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+            "@org_tensorflow//tensorflow/lite/kernels/internal:tensor",
+        ],
+)
+
+cc_test(
+    name = "optimized_encoder_test",
+    srcs = [
+        "optimized_encoder_test.cc",
+    ],
+    data = [
+        ":testdata",
+    ],
+    deps = [
+        ":double_array_trie_builder",
+        ":encoder_config",
+        ":model_converter",
+        ":optimized_encoder",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings:str_format",
+        "@com_google_googletest//:gtest_main",
+        "@com_google_sentencepiece//src:sentencepiece_cc_proto",
+        "@com_google_sentencepiece//src:sentencepiece_processor",
+        "@org_tensorflow//tensorflow/core:lib",
+    ],
+)
+
+cc_test(
+    name = "optimized_decoder_test",
+    srcs = [
+        "optimized_decoder_test.cc",
+    ],
+    data = [
+        ":testdata",
+    ],
+    deps = [
+        ":decoder_config",
+        ":double_array_trie_builder",
+        ":model_converter",
+        ":optimized_decoder",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings:str_format",
+        "@com_google_googletest//:gtest_main",
+        "@com_google_sentencepiece//src:sentencepiece_cc_proto",
+        "@com_google_sentencepiece//src:sentencepiece_processor",
+        "@org_tensorflow//tensorflow/core:lib",
+    ],
+)
+
+cc_library(
+    name = "py_tflite_registerer",
+    srcs = ["py_tflite_registerer.cc"],
+    hdrs = ["py_tflite_registerer.h"],
+    deps = [
+        ":sentencepiece_detokenizer_tflite",
+        ":sentencepiece_tokenizer_tflite",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+    ],
+    alwayslink = 1,
+)
+
+config_setting(
+    name = "armeabi_v7a_and_fastbuild",
+    values = {
+        "cpu": "armeabi-v7a",
+        "compilation_mode": "fastbuild",
+    },
+    visibility = ["//visibility:public"],
+)
+
+config_setting(
+    name = "armeabi_v7a_and_dbg",
+    values = {
+        "cpu": "armeabi-v7a",
+        "compilation_mode": "dbg",
+    },
+    visibility = ["//visibility:public"],
+)
+
+config_setting(
+    name = "android",
+    values = {"crosstool_top": "//external:android/crosstool"},
+    visibility = ["//visibility:public"],
+)
+
+config_setting(
+    name = "macos_i386",
+    values = {
+        "apple_platform_type": "macos",
+        "cpu": "darwin",
+    },
+    visibility = ["//visibility:public"],
+)
+
+config_setting(
+    name = "macos_x86_64",
+    values = {
+        "apple_platform_type": "macos",
+        "cpu": "darwin_x86_64",
+    },
+    visibility = ["//visibility:public"],
+)
+
+alias(
+    name = "macos",
+    actual = select({
+        ":macos_i386": ":macos_i386",
+        ":macos_x86_64": ":macos_x86_64",
+        "//conditions:default": ":macos_i386",  # Arbitrarily chosen from above.
+    }),
+    visibility = ["//visibility:public"],
+)
+
+config_setting(
+    name = "ios",
+    values = {
+        "crosstool_top": "@bazel_tools//tools/cpp:toolchain",
+        "apple_platform_type": "ios",
+    },
+    visibility = ["//visibility:public"],
+)
+
+alias(
+    name = "apple",
+    actual = select({
+        ":macos": ":macos",
+        ":ios": ":ios",
+        "//conditions:default": ":ios",  # Arbitrarily chosen from above.
+    }),
+    visibility = ["//visibility:public"],
+)
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/config.fbs b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/config.fbs
new file mode 100644
index 0000000..eba0bd8
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/config.fbs
@@ -0,0 +1,25 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+namespace tflite.ops.custom.sentencepiece;
+
+table Trie {
+  nodes: [uint32];
+}
+
+
+enum EncoderVersion: byte {
+  SENTENCE_PIECE = 0,
+}
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/decoder_config.fbs b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/decoder_config.fbs
new file mode 100644
index 0000000..4a230ed
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/decoder_config.fbs
@@ -0,0 +1,43 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+include "config.fbs";
+
+namespace tflite.ops.custom.sentencepiece;
+
+
+table DecoderConfig {
+  version: EncoderVersion = SENTENCE_PIECE;
+
+  // The offset for encoding, usually used when codes with low codes are reserved
+  // for some special needs.
+  encoding_offset: int32;
+
+  // A vector of strings that represent sentencepieces.
+  decode_pieces: [string];
+
+  // TODO(mgubin): Currently is not populated, haven't seen any Sentencepiece
+  // model with a denormalizer.
+  denormalized_prefixes: Trie;
+  denormalized_replacements: [byte];
+
+  // During encoding a dummy prefix (a whitespace) can be added to the input string,
+  // if this flag is true, this prefix will be removed.
+  remove_dummy_prefix: bool;
+
+}
+
+
+root_type DecoderConfig;
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie.h
new file mode 100644
index 0000000..547e0ea
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie.h
@@ -0,0 +1,120 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_DOUBLE_ARRAY_TRIE_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_DOUBLE_ARRAY_TRIE_H_
+
+#include <functional>
+#include <vector>
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/config_generated.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/utils.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+// A trie node specifies a node in the tree, either an intermediate node or
+// a leaf node.
+// A leaf node contains the id as an int of the string match. This id is encoded
+// in the lower 31 bits, thus the number of distinct ids is 2^31.
+// An intermediate node has an associated label and an offset to its children.
+// The label is encoded in the least significant byte and must match the input
+// character during matching.
+
+// A memory mappable trie, compatible with Darts::DoubleArray.
+class DoubleArrayTrie {
+ public:
+  struct Match {
+    Match() {}
+    Match(int id, int match_length) : id(id), match_length(match_length) {}
+    int id = -1;
+    int match_length = -1;
+    bool empty() const { return match_length == -1; }
+    bool operator==(const Match& m) const {
+      return m.id == id && m.match_length == match_length;
+    }
+  };
+
+  // nodes and nodes_length specify the array of the nodes of the trie.
+  explicit DoubleArrayTrie(const flatbuffers::Vector<uint32_t>* nodes)
+      : nodes_(nodes) {}
+
+  // Finds matches that are prefixes of a string.
+  template <typename callback>
+  void IteratePrefixMatches(const utils::string_view& input,
+                            callback update_fn) const;
+
+  // Finds the longest prefix match of a string.
+  Match LongestPrefixMatch(const utils::string_view& input) const {
+    Match match;
+    IteratePrefixMatches(input, [&match](const Match& m) { match = m; });
+    return match;
+  }
+
+ private:
+  // Returns whether a node as a leaf as a child.
+  bool has_leaf(uint32_t i) const { return ((*nodes_)[i]) & 0x100; }
+
+  // Returns a value associated with a node. Available when a node is a leaf.
+  int value(uint32_t i) const {
+    return static_cast<int>(((*nodes_)[i]) & 0x7fffffff);
+  }
+
+  // Returns a label associated with a node.
+  // A leaf node will have the MSB set and thus return an invalid label.
+  int32_t label(uint32_t i) const { return ((*nodes_)[i]) & 0x800000ff; }
+
+  // Returns offset to children.
+  int32_t offset(uint32_t i) const {
+    const uint32_t node = (*nodes_)[i];
+    return (node >> 10) << ((node & 0x200) >> 6);
+  }
+
+  const flatbuffers::Vector<uint32_t>* nodes_;
+};
+
+template <typename callback>
+void DoubleArrayTrie::IteratePrefixMatches(const utils::string_view& input,
+                                           callback update_fn) const {
+  if (nodes_->size() == 0) {
+    return;
+  }
+  uint32_t pos = offset(0);
+  for (int i = 0; i < input.length(); ++i) {
+    pos ^= static_cast<unsigned char>(input.at(i));
+    if (pos < 0 || pos >= nodes_->size() || label(pos) != input.at(i)) {
+      // No match, exit.
+      return;
+    }
+    const bool node_has_leaf = has_leaf(pos);
+    pos ^= offset(pos);
+    if (pos < 0 || pos >= nodes_->size()) {
+      // We can get here only if the structure is corrupted.
+      return;
+    }
+    if (node_has_leaf) {
+      update_fn(Match(value(pos), i + 1));
+    }
+  }
+}
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_DOUBLE_ARRAY_TRIE_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.cc
new file mode 100644
index 0000000..72b7262
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.cc
@@ -0,0 +1,81 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.h"
+
+#include <algorithm>
+#include <memory>
+
+#include "include/darts.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+std::vector<uint32_t> BuildTrie(const std::vector<std::string>& data) {
+  std::vector<int> ids;
+  ids.reserve(data.size());
+  for (int i = 0; i < data.size(); ++i) {
+    ids.push_back(i);
+  }
+  return BuildTrie(data, ids);
+}
+
+std::vector<uint32_t> BuildTrie(const std::vector<std::string>& data,
+                                const std::vector<int>& ids) {
+  // We make strong assumptions about binary structure of trie.
+  struct OneElement {
+    OneElement(const std::string* key_, int index_)
+        : key(key_), index(index_) {}
+    const std::string* key;
+    int index;
+    bool operator<(const OneElement& el) const { return *key < *el.key; }
+  };
+  std::vector<OneElement> elements;
+  elements.reserve(data.size());
+  auto data_iterator = std::begin(data);
+  auto ids_iterator = std::begin(ids);
+  for (; data_iterator != std::end(data) && ids_iterator != std::end(ids);
+       ++data_iterator, ++ids_iterator) {
+    elements.emplace_back(&(*data_iterator), *ids_iterator);
+  }
+  // Sort by keys.
+  std::sort(elements.begin(), elements.end());
+
+  // Create vectors to build the trie.
+  std::vector<const char*> strings;
+  std::vector<int32_t> indexes;
+  strings.reserve(data.size());
+  indexes.reserve(data.size());
+  for (const auto& el : elements) {
+    strings.push_back(el.key->c_str());
+    indexes.push_back(el.index);
+  }
+  auto trie = std::make_unique<Darts::DoubleArray>();
+  trie->build(data.size(), const_cast<char**>(&strings[0]), nullptr,
+              &indexes[0]);
+  // We make strong assumptions about internal Darts trie structure:
+  // - it is a vector of 32 bit signed integers
+  // - the "array" is the only one structure that contains all information about
+  // the trie.
+  const uint32_t* trie_data = static_cast<const uint32_t*>(trie->array());
+  return std::vector<uint32_t>(trie_data, trie_data + trie->size());
+}
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.h
new file mode 100644
index 0000000..bc618ab
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.h
@@ -0,0 +1,41 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_DOUBLE_ARRAY_TRIE_BUILDER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_DOUBLE_ARRAY_TRIE_BUILDER_H_
+
+#include <string>
+#include <vector>
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/config_generated.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/utils.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+std::vector<uint32_t> BuildTrie(const std::vector<std::string>& data,
+                                const std::vector<int>& ids);
+
+// A variant where ids are indexes in data.
+std::vector<uint32_t> BuildTrie(const std::vector<std::string>& data);
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_DOUBLE_ARRAY_TRIE_BUILDER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_test.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_test.cc
new file mode 100644
index 0000000..8a53d09
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_test.cc
@@ -0,0 +1,78 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie.h"
+
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config_generated.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+TEST(DoubleArrayTrieTest, Match) {
+  flatbuffers::FlatBufferBuilder builder(1024);
+  const std::vector<std::string> test_strings = {"A", "AAX", "AA", "B"};
+  const auto trie_vector = builder.CreateVector(BuildTrie(test_strings));
+  TrieBuilder trie_builder(builder);
+  trie_builder.add_nodes(trie_vector);
+  const auto pieces = trie_builder.Finish();
+  EncoderConfigBuilder ecb(builder);
+  ecb.add_pieces(pieces);
+  FinishEncoderConfigBuffer(builder, ecb.Finish());
+  const EncoderConfig* config = GetEncoderConfig(builder.GetBufferPointer());
+  DoubleArrayTrie dat(config->pieces()->nodes());
+  EXPECT_EQ(dat.LongestPrefixMatch(utils::string_view("AAL")),
+            DoubleArrayTrie::Match(2, 2));
+
+  std::vector<DoubleArrayTrie::Match> matches;
+  dat.IteratePrefixMatches(
+      utils::string_view("AAXL"),
+      [&matches](const DoubleArrayTrie::Match& m) { matches.push_back(m); });
+  EXPECT_THAT(matches, testing::ElementsAre(DoubleArrayTrie::Match(0, 1),
+                                            DoubleArrayTrie::Match(2, 2),
+                                            DoubleArrayTrie::Match(1, 3)));
+}
+
+TEST(DoubleArrayTrieTest, ComplexMatch) {
+  flatbuffers::FlatBufferBuilder builder(1024);
+  const std::vector<std::string> test_strings = {"\xe2\x96\x81the", ",", "s",
+                                                 "\xe2\x96\x81Hello"};
+  const std::vector<int> test_ids = {0, 5, 10, 15};
+  const auto trie_vector =
+      builder.CreateVector(BuildTrie(test_strings, test_ids));
+  TrieBuilder trie_builder(builder);
+  trie_builder.add_nodes(trie_vector);
+  const auto pieces = trie_builder.Finish();
+  EncoderConfigBuilder ecb(builder);
+  ecb.add_pieces(pieces);
+  FinishEncoderConfigBuffer(builder, ecb.Finish());
+  const EncoderConfig* config = GetEncoderConfig(builder.GetBufferPointer());
+  DoubleArrayTrie dat(config->pieces()->nodes());
+
+  std::vector<DoubleArrayTrie::Match> matches;
+  dat.IteratePrefixMatches(
+      utils::string_view("\xe2\x96\x81Hello"),
+      [&matches](const DoubleArrayTrie::Match& m) { matches.push_back(m); });
+  EXPECT_THAT(matches, testing::ElementsAre(DoubleArrayTrie::Match(15, 8)));
+}
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config.fbs b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config.fbs
new file mode 100644
index 0000000..7f1f2ba
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config.fbs
@@ -0,0 +1,52 @@
+// Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+include "config.fbs";
+
+namespace tflite.ops.custom.sentencepiece;
+
+table EncoderConfig {
+  // Version of the encoder.
+  version: EncoderVersion = SENTENCE_PIECE;
+  start_code: int32 = 0;
+  end_code: int32 = 0;
+
+  unknown_code: int32 = -1;
+  // Weight of "unknown code" when encoding. "Penalty" because it usually has a
+  // big negative weight,less than any other sentencepiece.
+  unknown_penalty: float = 0;
+
+  // The offset for encoding, usually used when codes with low codes are reserved
+  // for some special needs.
+  encoding_offset: int32;
+
+  // String pieces for encoding.
+  pieces: Trie;
+  pieces_scores: [float];
+
+  // Normalization related parameters.
+  remove_extra_whitespaces: bool;
+
+  // Add a whitespace prefix before encoding.
+  add_dummy_prefix: bool;
+
+  // Escape whitespaces during encoding so the decoder can restore them exactly as
+  // in the input.
+  escape_whitespaces: bool;
+
+  // Normalization parameters.
+  normalized_prefixes: Trie;
+  normalized_replacements: [byte];
+}
+
+root_type EncoderConfig;
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.cc
new file mode 100644
index 0000000..73e853f
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.cc
@@ -0,0 +1,197 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.h"
+
+#include "absl/status/status.h"
+#include "absl/strings/str_replace.h"
+#include "src/sentencepiece_model.pb.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/decoder_config_generated.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config_generated.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_constants.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+std::tuple<std::vector<uint32_t>, std::vector<int8_t>>
+DecodePrecompiledCharsmap(
+    const ::sentencepiece::NormalizerSpec& normalizer_spec) {
+  // This function "undoes" encoding done by
+  // sentencepiece::normalizer::Normalizer::EncodePrecompiledCharsMap.
+  const char* precompiled_map = normalizer_spec.precompiled_charsmap().data();
+  const uint32_t trie_size =
+      *reinterpret_cast<const uint32_t*>(precompiled_map);
+  const uint32_t* trie_ptr =
+      reinterpret_cast<const uint32_t*>(precompiled_map + sizeof(uint32_t));
+  const int8_t* normalized_ptr = reinterpret_cast<const int8_t*>(
+      precompiled_map + sizeof(uint32_t) + trie_size);
+  const int normalized_size = normalizer_spec.precompiled_charsmap().length() -
+                              sizeof(uint32_t) - trie_size;
+  return std::make_tuple(
+      std::vector<uint32_t>(trie_ptr, trie_ptr + trie_size / sizeof(uint32_t)),
+      std::vector<int8_t>(normalized_ptr, normalized_ptr + normalized_size));
+}
+
+tflite::support::StatusOr<std::string> ConvertSentencepieceModelToFlatBuffer(
+    const std::string& model_config_str, int encoding_offset) {
+  ::sentencepiece::ModelProto model_config;
+  if (!model_config.ParseFromString(model_config_str)) {
+    return absl::InvalidArgumentError(
+        "Invalid configuration, can't parse SentencePiece model config " +
+        model_config.InitializationErrorString());
+  }
+  // Convert sentencepieces.
+  std::vector<std::string> pieces;
+  pieces.reserve(model_config.pieces_size());
+  std::vector<float> scores;
+  scores.reserve(model_config.pieces_size());
+  std::vector<int> ids;
+  ids.reserve(model_config.pieces_size());
+  float min_score = 0.0;
+  int index = 0;
+  for (const auto& piece : model_config.pieces()) {
+    switch (piece.type()) {
+      case ::sentencepiece::ModelProto::SentencePiece::NORMAL:
+      case ::sentencepiece::ModelProto::SentencePiece::USER_DEFINED:
+        pieces.push_back(piece.piece());
+        ids.push_back(index);
+        if (piece.score() < min_score) {
+          min_score = piece.score();
+        }
+        break;
+      case ::sentencepiece::ModelProto::SentencePiece::UNKNOWN:
+      case ::sentencepiece::ModelProto::SentencePiece::CONTROL:
+        // Ignore unknown and control codes.
+        break;
+      default:
+        return absl::InvalidArgumentError("Invalid SentencePiece piece type " +
+                                          piece.piece());
+    }
+    scores.push_back(piece.score());
+    ++index;
+  }
+  flatbuffers::FlatBufferBuilder builder(1024);
+  const auto pieces_trie_vector = builder.CreateVector(BuildTrie(pieces, ids));
+  const auto pieces_score_vector = builder.CreateVector(scores);
+  TrieBuilder pieces_trie_builder(builder);
+  pieces_trie_builder.add_nodes(pieces_trie_vector);
+  const auto pieces_trie_fbs = pieces_trie_builder.Finish();
+
+  // Converting normalization.
+  const auto [normalization_trie, normalization_strings] =
+      DecodePrecompiledCharsmap(model_config.normalizer_spec());
+  const auto normalization_trie_vector =
+      builder.CreateVector(normalization_trie);
+  TrieBuilder normalization_trie_builder(builder);
+  normalization_trie_builder.add_nodes(normalization_trie_vector);
+  const auto normalization_trie_fbs = normalization_trie_builder.Finish();
+  const auto normalization_strings_fbs =
+      builder.CreateVector(normalization_strings);
+
+  EncoderConfigBuilder ecb(builder);
+  ecb.add_version(EncoderVersion::EncoderVersion_SENTENCE_PIECE);
+  ecb.add_start_code(model_config.trainer_spec().bos_id());
+  ecb.add_end_code(model_config.trainer_spec().eos_id());
+  ecb.add_unknown_code(model_config.trainer_spec().unk_id());
+  ecb.add_unknown_penalty(min_score - kUnkPenalty);
+  ecb.add_encoding_offset(encoding_offset);
+  ecb.add_pieces(pieces_trie_fbs);
+  ecb.add_pieces_scores(pieces_score_vector);
+  ecb.add_remove_extra_whitespaces(
+      model_config.normalizer_spec().remove_extra_whitespaces());
+  ecb.add_add_dummy_prefix(model_config.normalizer_spec().add_dummy_prefix());
+  ecb.add_escape_whitespaces(
+      model_config.normalizer_spec().escape_whitespaces());
+  ecb.add_normalized_prefixes(normalization_trie_fbs);
+  ecb.add_normalized_replacements(normalization_strings_fbs);
+  FinishEncoderConfigBuffer(builder, ecb.Finish());
+  return std::string(reinterpret_cast<const char*>(builder.GetBufferPointer()),
+                     builder.GetSize());
+}
+
+tflite::support::StatusOr<std::string>
+ConvertSentencepieceModelToFlatBufferForDecoder(
+    const std::string& model_config_str, int encoding_offset) {
+  ::sentencepiece::ModelProto model_config;
+  if (!model_config.ParseFromString(model_config_str)) {
+    return absl::InvalidArgumentError(
+        "Invalid configuration, can't parse SentencePiece model config " +
+        model_config.InitializationErrorString());
+  }
+  flatbuffers::FlatBufferBuilder builder(1024);
+  // Collect sentencepieces.
+  std::vector<std::string> pieces;
+  for (const auto& piece : model_config.pieces()) {
+    // In the original library all pieces processing is done during decoding.
+    // Because it is independent from context or parameters we can do it in
+    // advance here.
+    switch (piece.type()) {
+      case ::sentencepiece::ModelProto::SentencePiece::NORMAL:
+      case ::sentencepiece::ModelProto::SentencePiece::USER_DEFINED:
+        pieces.push_back(
+            absl::StrReplaceAll(piece.piece(), {{kSpaceSymbol, " "}}));
+        break;
+      case ::sentencepiece::ModelProto::SentencePiece::UNKNOWN:
+        pieces.push_back(
+            kDefaultUnknownSymbol);  // Always decode with the default unknown.
+        break;
+      default:
+        pieces.push_back("");
+    }
+  }
+  const auto pieces_fbs = builder.CreateVectorOfStrings(pieces);
+  DecoderConfigBuilder decb(builder);
+
+  decb.add_version(EncoderVersion::EncoderVersion_SENTENCE_PIECE);
+  decb.add_encoding_offset(encoding_offset);
+  decb.add_decode_pieces(pieces_fbs);
+  decb.add_remove_dummy_prefix(
+      model_config.normalizer_spec().add_dummy_prefix());
+
+  FinishDecoderConfigBuffer(builder, decb.Finish());
+  return std::string(reinterpret_cast<const char*>(builder.GetBufferPointer()),
+                     builder.GetSize());
+}
+
+int GetVocabularySize(const std::string& model_string) {
+  const EncoderConfig* config = GetEncoderConfig(model_string.data());
+  return config->pieces_scores()->size() + config->encoding_offset();
+}
+
+std::string ConvertSentencepieceModel(const std::string& model_string) {
+  const auto result = ConvertSentencepieceModelToFlatBuffer(model_string);
+  // TODO(mgubin): Propogate error to the Python code and throw correct
+  // exception.
+  assert(result.status().ok());
+  return result.value();
+}
+
+std::string ConvertSentencepieceModelForDecoder(
+    const std::string& model_string) {
+  const auto result =
+      ConvertSentencepieceModelToFlatBufferForDecoder(model_string);
+  // TODO(mgubin): Propogate error to the Python code and throw correct
+  // exception.
+  assert(result.status().ok());
+  return result.value();
+}
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.h
new file mode 100644
index 0000000..5687b62
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.h
@@ -0,0 +1,52 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_MODEL_CONVERTER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_MODEL_CONVERTER_H_
+#include <string>
+
+#include "tensorflow_lite_support/cc/port/statusor.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+// Converts Sentencepiece configuration to flatbuffer format.
+// encoding_offset is used by some encoders that combine different encodings.
+tflite::support::StatusOr<std::string> ConvertSentencepieceModelToFlatBuffer(
+    const std::string& model_config_str, int encoding_offset = 0);
+
+// Converts Sentencepiece configuration to flatbuffer format for encoder.
+// encoding_offset is used by some encoders that combine different encodings.
+tflite::support::StatusOr<std::string>
+ConvertSentencepieceModelToFlatBufferForDecoder(
+    const std::string& model_config_str, int encoding_offset = 0);
+
+// The functions that are provided for the Python wrapper.
+std::string ConvertSentencepieceModel(const std::string& model_string);
+std::string ConvertSentencepieceModelForDecoder(
+    const std::string& model_string);
+
+// Returns size of a vocabulary from Sentencepiece configuration in flatbuffer
+// format.
+int GetVocabularySize(const std::string& model_string);
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_MODEL_CONVERTER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/native.bzl b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/native.bzl
new file mode 100644
index 0000000..87695a4
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/native.bzl
@@ -0,0 +1,86 @@
+"""Build definitions supporting platform-independent native build."""
+
+load("@org_tensorflow//tensorflow:tensorflow.bzl", "tf_copts", "tf_opts_nortti_if_android")
+load("@bazel_skylib//lib:selects.bzl", "selects")
+
+def micore_if(android, ios = [], default = []):
+    """Helper to create a select.
+
+    Args:
+      android: what to return if compiling for Android.
+      ios: what to return if compiling for iOS.
+      default: what to return otherwise.
+    Returns:
+      the `android` list for Android compilation and the
+      `default` list otherwise.
+    """
+    return select({
+        ":android": android,
+        ":apple": ios,
+        "//conditions:default": default,
+    })
+
+def micore_tf_copts():
+    """C options for Tensorflow builds.
+
+    Returns:
+      a list of copts which must be used by each cc_library which
+      refers to Tensorflow. Enables the library to compile both for
+      Android and for Linux.
+    """
+    return tf_copts(android_optimization_level_override = None) + tf_opts_nortti_if_android() + [
+        "-Wno-narrowing",
+        "-Wno-sign-compare",
+        "-Wno-overloaded-virtual",
+    ] + micore_if(
+        android = [
+            # Set a define so Tensorflow's register_types.h
+            # adopts to support a rich set of types, to be pruned by
+            # selective registration.
+            "-DSUPPORT_SELECTIVE_REGISTRATION",
+            # Selective registration uses constexprs with recursive
+            # string comparisons; that can lead to compiler errors, so
+            # we increase the constexpr recursion depth.
+            "-fconstexpr-depth=1024",
+        ],
+    ) + selects.with_or({
+        # If building for armeabi-v7a, and if compilation_mode is 'fastbuild'
+        # or 'dbg' then forcefully add -Oz to the list compiler options.
+        # Without it, some TF dependencies can't build (b/112286436). If
+        # compilation_mode is 'opt' then rely on the toolchain default.
+        (
+            ":armeabi_v7a_and_fastbuild",
+            ":armeabi_v7a_and_dbg",
+        ): ["-Oz"],
+        "//conditions:default": [],
+    })
+
+def micore_tf_deps():
+    """Dependencies for Tensorflow builds.
+
+    Returns:
+      list of dependencies which must be used by each cc_library
+      which refers to Tensorflow. Enables the library to compile both for
+      Android and for Linux. Use this macro instead of directly
+      declaring dependencies on Tensorflow.
+    """
+    return micore_if(
+        android = [
+            # Link to library which does not contain any ops.
+            "@org_tensorflow//tensorflow/core:portable_tensorflow_lib_lite",
+            "@gemmlowp//:eight_bit_int_gemm",
+            "@fft2d//:fft2d",
+        ],
+        ios = [
+            "@org_tensorflow//tensorflow/core:portable_tensorflow_lib",
+            "@gemmlowp//:eight_bit_int_gemm",
+            "@fft2d//:fft2d",
+        ],
+        default = [
+            # Standard references for Tensorflow when building for Linux. We use
+            # an indirection via the alias targets below, to facilitate whitelisting
+            # these deps in the mobile license presubmit checks.
+            "@local_config_tf//:libtensorflow_framework",
+            "@local_config_tf//:tf_header_lib",
+        ],
+    )
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.cc
new file mode 100644
index 0000000..86e186d
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.cc
@@ -0,0 +1,63 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.h"
+
+#include <string>
+#include <tuple>
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/decoder_config_generated.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+DecoderResult DecodeString(const std::vector<int>& encoded,
+                           const void* config_buffer) {
+  DecoderResult result;
+
+  // Get the config from the buffer.
+  const DecoderConfig* config = GetDecoderConfig(config_buffer);
+  if (config->version() != EncoderVersion::EncoderVersion_SENTENCE_PIECE) {
+    result.type = DecoderResultType::WRONG_CONFIG;
+    return result;
+  }
+  bool remove_dummy_prefix = config->remove_dummy_prefix();
+  const auto config_pieces = config->decode_pieces();
+  for (const auto code : encoded) {
+    const int real_code = code - config->encoding_offset();
+    if (real_code >= config_pieces->size()) {
+      result.type = DecoderResultType::INVALID_INPUT;
+      return result;
+    }
+    const auto& piece_text = config_pieces->GetAsString(real_code);
+    const char* piece_str = piece_text->c_str();
+    if (remove_dummy_prefix && *piece_str == ' ') {
+      ++piece_str;
+    }
+    result.decoded.append(piece_str);
+    remove_dummy_prefix = false;
+  }
+  // TODO(mgubin): Denormalize the string, haven't seen any Sentencepiece model
+  // with a denormalizer.
+  return result;
+}
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.h
new file mode 100644
index 0000000..a442468
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.h
@@ -0,0 +1,50 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_OPTIMIZED_DECODER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_OPTIMIZED_DECODER_H_
+
+// Sentencepiece decoder optimized with memmapped model.
+
+#include <string>
+#include <vector>
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+enum class DecoderResultType {
+  SUCCESS = 0,
+  WRONG_CONFIG = 1,
+  INVALID_INPUT = 2
+};
+
+struct DecoderResult {
+  DecoderResultType type = DecoderResultType::SUCCESS;
+  std::string decoded;
+};
+
+// Decodes one string from a vector of id. Takes the configuration as a
+// type-erased  buffer.
+DecoderResult DecodeString(const std::vector<int>& encoded,
+                           const void* config_buffer);
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_OPTIMIZED_DECODER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder_test.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder_test.cc
new file mode 100644
index 0000000..04d1c85
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder_test.cc
@@ -0,0 +1,90 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.h"
+
+#include <fstream>
+
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include "src/sentencepiece.pb.h"
+#include "src/sentencepiece_processor.h"
+#include "tensorflow/core/platform/env.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+namespace internal {
+
+tensorflow::Status TFReadFileToString(const std::string& filepath,
+                                      std::string* data) {
+  return tensorflow::ReadFileToString(tensorflow::Env::Default(),
+                                      /*test_path*/ filepath, data);
+}
+
+absl::Status StdReadFileToString(const std::string& filepath,
+                                 std::string* data) {
+  std::ifstream infile(filepath);
+  if (!infile.is_open()) {
+    return absl::NotFoundError(
+        absl::StrFormat("Error when opening %s", filepath));
+  }
+  std::string contents((std::istreambuf_iterator<char>(infile)),
+                       (std::istreambuf_iterator<char>()));
+  data->append(contents);
+  infile.close();
+  return absl::OkStatus();
+}
+
+}  // namespace internal
+
+namespace {
+static char kConfigFilePath[] =
+    "tensorflow_lite_support/custom_ops/kernel/"
+    "sentencepiece/testdata/sentencepiece.model";
+
+TEST(OptimizedEncoder, ConfigConverter) {
+  std::string config;
+  auto status = internal::StdReadFileToString(kConfigFilePath, &config);
+
+  ASSERT_TRUE(status.ok());
+
+  ::sentencepiece::SentencePieceProcessor processor;
+  ASSERT_OK(processor.LoadFromSerializedProto(config));
+  const auto converted_model = ConvertSentencepieceModelForDecoder(config);
+  const std::string test_string("Hello world!\\xF0\\x9F\\x8D\\x95");
+  ::sentencepiece::SentencePieceText reference_encoded;
+  CHECK_OK(processor.Encode(test_string, &reference_encoded));
+
+  std::vector<int> encoded_vector;
+  encoded_vector.reserve(reference_encoded.pieces_size());
+  for (const auto& piece : reference_encoded.pieces()) {
+    encoded_vector.push_back(piece.id());
+  }
+  std::string ref_decoded;
+  ASSERT_OK(processor.Decode(encoded_vector, &ref_decoded));
+  const auto decoded = DecodeString(encoded_vector, converted_model.data());
+  ASSERT_EQ(decoded.type, DecoderResultType::SUCCESS);
+  ASSERT_EQ(ref_decoded, decoded.decoded);
+}
+}  // namespace
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.cc
new file mode 100644
index 0000000..5a59ee4
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.cc
@@ -0,0 +1,239 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.h"
+
+#include <algorithm>
+#include <tuple>
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config_generated.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+namespace {
+
+const char kSpaceSymbol[] = "\xe2\x96\x81";
+
+template <typename processing_callback>
+std::tuple<std::string, std::vector<int>> process_string(
+    const std::string& input, const std::vector<int>& offsets,
+    const processing_callback& pc) {
+  std::string result_string;
+  result_string.reserve(input.size());
+  std::vector<int> result_offsets;
+  result_offsets.reserve(offsets.size());
+  for (int i = 0, j = 0; i < input.size();) {
+    auto [consumed, new_string] = pc(input.data() + i, input.size() - i);
+    if (consumed == 0) {
+      // Skip the current byte and move forward.
+      result_string.push_back(input[i]);
+      result_offsets.push_back(offsets[j]);
+      i++;
+      j++;
+      continue;
+    }
+    result_string.append(new_string.data(), new_string.length());
+    for (int i = 0; i < new_string.length(); ++i) {
+      result_offsets.push_back(offsets[j]);
+    }
+    j += consumed;
+    i += consumed;
+  }
+  return std::make_tuple(result_string, result_offsets);
+}
+
+inline char is_whitespace(char c) {
+  return c == ' ' || c == '\t' || c == '\r' || c == '\n';
+}
+
+std::tuple<int, utils::string_view> remove_extra_whitespaces(const char* data,
+                                                             int len) {
+  if (len == 0 || !is_whitespace(*data)) {
+    return std::make_tuple(0, utils::string_view(nullptr, 0));
+  }
+  int num_consumed = 1;
+  for (; num_consumed < len && is_whitespace(data[num_consumed]);
+       ++num_consumed) {
+  }
+  return num_consumed > 1
+             ? std::make_tuple(num_consumed, utils::string_view(" ", 1))
+             : std::make_tuple(0, utils::string_view(nullptr, 0));
+}
+
+std::tuple<int, utils::string_view> find_replacement(
+    const char* data, int len, const DoubleArrayTrie& dat,
+    const flatbuffers::Vector<int8_t>& replacements) {
+  const auto max_match = dat.LongestPrefixMatch(utils::string_view(data, len));
+  if (!max_match.empty()) {
+    // Because flatbuffer byte is signed char which is not the same as char,
+    // there is the reinterpret_cast here.
+    const char* replaced_string_ptr =
+        reinterpret_cast<const char*>(replacements.data() + max_match.id);
+    return std::make_tuple(max_match.match_length,
+                           utils::string_view(replaced_string_ptr));
+  }
+  return std::make_tuple(0, utils::string_view(nullptr, 0));
+}
+}  // namespace
+
+std::tuple<std::string, std::vector<int>> NormalizeString(
+    const std::string& in_string, const EncoderConfig& config) {
+  std::vector<int> output_offsets;
+  std::string result = in_string;
+  output_offsets.reserve(in_string.length());
+  for (int i = 0; i < in_string.length(); ++i) {
+    output_offsets.push_back(i);
+  }
+  if (in_string.empty()) {
+    return std::make_tuple(result, output_offsets);
+  }
+  if (config.add_dummy_prefix()) {
+    result.insert(result.begin(), ' ');
+    output_offsets.insert(output_offsets.begin(), 0);
+  }
+  // Greedely replace normalized_prefixes with normalized_replacements
+  if (config.normalized_prefixes() != nullptr &&
+      config.normalized_replacements() != nullptr) {
+    const DoubleArrayTrie normalized_prefixes_matcher(
+        config.normalized_prefixes()->nodes());
+    const auto norm_replace = [&config, &normalized_prefixes_matcher](
+                                  const char* data, int len) {
+      return find_replacement(data, len, normalized_prefixes_matcher,
+                              *config.normalized_replacements());
+    };
+    std::tie(result, output_offsets) =
+        process_string(result, output_offsets, norm_replace);
+  }
+  if (config.remove_extra_whitespaces()) {
+    std::tie(result, output_offsets) =
+        process_string(result, output_offsets, remove_extra_whitespaces);
+    if (!result.empty() && is_whitespace(result.back())) {
+      result.pop_back();
+      output_offsets.pop_back();
+    }
+  }
+  if (config.escape_whitespaces()) {
+    const auto replace_whitespaces = [](const char* data, int len) {
+      if (len > 0 && is_whitespace(*data)) {
+        return std::make_tuple(1, utils::string_view(kSpaceSymbol));
+      }
+      return std::make_tuple(0, utils::string_view(nullptr, 0));
+    };
+    std::tie(result, output_offsets) =
+        process_string(result, output_offsets, replace_whitespaces);
+  }
+
+  return std::make_tuple(result, output_offsets);
+}
+
+EncoderResult EncodeNormalizedString(const std::string& str,
+                                     const std::vector<int>& offsets,
+                                     const EncoderConfig& config, bool add_bos,
+                                     bool add_eos, bool reverse) {
+  const DoubleArrayTrie piece_matcher(config.pieces()->nodes());
+  const flatbuffers::Vector<float>* piece_scores = config.pieces_scores();
+  const int unknown_code = config.unknown_code();
+  const float unknown_penalty = config.unknown_penalty();
+  struct LatticeElement {
+    float score = 0;
+    int code = -1;
+    int prev_position = -1;
+    LatticeElement(float score_, int code_, int prev_position_)
+        : score(score_), code(code_), prev_position(prev_position_) {}
+    LatticeElement() {}
+  };
+  const int length = str.length();
+  std::vector<LatticeElement> lattice(length + 1);
+  for (int i = 0; i < length; ++i) {
+    if (i > 0 && lattice[i].prev_position < 0) {
+      // This state is unreachable.
+      continue;
+    }
+    if (unknown_code >= 0) {
+      // Put unknown code.
+      const float penalized_score = lattice[i].score + unknown_penalty;
+      const int pos = i + 1;
+      LatticeElement& current_element = lattice[pos];
+      if (current_element.prev_position < 0 ||
+          current_element.score < penalized_score) {
+        current_element = LatticeElement(
+            penalized_score, unknown_code,
+            // If the current state is already reached by unknown code, merge
+            // states.
+            lattice[i].code == unknown_code ? lattice[i].prev_position : i);
+      }
+    }
+    auto lattice_update = [&lattice, i,
+                           piece_scores](const DoubleArrayTrie::Match& m) {
+      LatticeElement& target_element = lattice[i + m.match_length];
+      const float score = lattice[i].score + (*piece_scores)[m.id];
+      if (target_element.prev_position < 0 || target_element.score < score) {
+        target_element = LatticeElement(score, m.id, i);
+      }
+    };
+    piece_matcher.IteratePrefixMatches(
+        utils::string_view(str.data() + i, length - i), lattice_update);
+  }
+
+  EncoderResult result;
+  if (add_eos) {
+    result.codes.push_back(config.end_code());
+    result.offsets.push_back(length);
+  }
+  if (lattice[length].prev_position >= 0) {
+    for (int pos = length; pos > 0;) {
+      auto code = lattice[pos].code;
+      if (code != config.unknown_code()) {
+        code += config.encoding_offset();
+      }
+      result.codes.push_back(code);
+      pos = lattice[pos].prev_position;
+      result.offsets.push_back(offsets[pos]);
+    }
+  }
+  if (add_bos) {
+    result.codes.push_back(config.start_code());
+    result.offsets.push_back(0);
+  }
+  if (!reverse) {
+    std::reverse(result.codes.begin(), result.codes.end());
+    std::reverse(result.offsets.begin(), result.offsets.end());
+  }
+  return result;
+}
+
+EncoderResult EncodeString(const std::string& string, const void* config_buffer,
+                           bool add_bos, bool add_eos, bool reverse) {
+  // Get the config from the buffer.
+  const EncoderConfig* config = GetEncoderConfig(config_buffer);
+  if (config->version() != EncoderVersion::EncoderVersion_SENTENCE_PIECE) {
+    EncoderResult result;
+    result.type = EncoderResultType::WRONG_CONFIG;
+    return result;
+  }
+  std::string normalized_string;
+  std::vector<int> offsets;
+  std::tie(normalized_string, offsets) = NormalizeString(string, *config);
+  return EncodeNormalizedString(normalized_string, offsets, *config, add_bos,
+                                add_eos, reverse);
+}
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.h
new file mode 100644
index 0000000..44d6e88
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.h
@@ -0,0 +1,52 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_OPTIMIZED_ENCODER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_OPTIMIZED_ENCODER_H_
+
+// Sentencepiece encoder optimized with memmapped model.
+
+#include <string>
+#include <tuple>
+#include <vector>
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config_generated.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+enum class EncoderResultType { SUCCESS = 0, WRONG_CONFIG = 1 };
+
+struct EncoderResult {
+  EncoderResultType type = EncoderResultType::SUCCESS;
+  std::vector<int> codes;
+  std::vector<int> offsets;
+};
+std::tuple<std::string, std::vector<int>> NormalizeString(
+    const std::string& in_string, const EncoderConfig& config);
+
+// Encodes one string and returns ids and offsets. Takes the configuration as a
+// type-erased buffer.
+EncoderResult EncodeString(const std::string& string, const void* config_buffer,
+                           bool add_bos, bool add_eos, bool reverse);
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_OPTIMIZED_ENCODER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder_test.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder_test.cc
new file mode 100644
index 0000000..ad3cd27
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder_test.cc
@@ -0,0 +1,167 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.h"
+
+#include <fstream>
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/double_array_trie_builder.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/encoder_config_generated.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/model_converter.h"
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "src/sentencepiece.pb.h"
+#include "src/sentencepiece_processor.h"
+#include "tensorflow/core/platform/env.h"
+
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+namespace internal {
+
+tensorflow::Status TFReadFileToString(
+    const std::string& filepath, std::string* data) {
+  return tensorflow::ReadFileToString(
+      tensorflow::Env::Default(), /*test_path*/ filepath, data);
+}
+
+absl::Status StdReadFileToString(
+    const std::string& filepath, std::string* data) {
+  std::ifstream infile(filepath);
+  if (!infile.is_open()) {
+    return absl::NotFoundError(
+        absl::StrFormat("Error when opening %s", filepath));
+  }
+  std::string contents((std::istreambuf_iterator<char>(infile)),
+                       (std::istreambuf_iterator<char>()));
+  data->append(contents);
+  infile.close();
+  return absl::OkStatus();
+}
+}  // namespace internal
+
+namespace {
+
+static char kConfigFilePath[] =
+    "tensorflow_lite_support/custom_ops/kernel/"
+    "sentencepiece/testdata/sentencepiece.model";
+
+TEST(OptimizedEncoder, NormalizeStringWhitestpaces) {
+  flatbuffers::FlatBufferBuilder builder(1024);
+  EncoderConfigBuilder ecb(builder);
+  ecb.add_remove_extra_whitespaces(true);
+  ecb.add_add_dummy_prefix(true);
+  ecb.add_escape_whitespaces(true);
+  FinishEncoderConfigBuffer(builder, ecb.Finish());
+  const EncoderConfig* config = GetEncoderConfig(builder.GetBufferPointer());
+  {
+    const auto [res_string, offsets] = NormalizeString("x  y", *config);
+    EXPECT_EQ(res_string, "\xe2\x96\x81x\xe2\x96\x81y");
+    EXPECT_THAT(offsets, ::testing::ElementsAre(0, 0, 0, 0, 1, 1, 1, 3));
+  }
+  {
+    const auto [res_string, offsets] = NormalizeString("\tx  y\n", *config);
+    EXPECT_EQ(res_string, "\xe2\x96\x81x\xe2\x96\x81y");
+    EXPECT_THAT(offsets, ::testing::ElementsAre(0, 0, 0, 1, 2, 2, 2, 4));
+  }
+}
+
+TEST(OptimizedEncoder, NormalizeStringReplacement) {
+  flatbuffers::FlatBufferBuilder builder(1024);
+  const std::vector<std::string> norm_prefixes = {"A", "AA", "AAA", "AAAA"};
+  const char norm_replacements[] = "A1\0A2\0A3\0A4";
+  const auto trie_vector =
+      builder.CreateVector(BuildTrie(norm_prefixes, {0, 3, 6, 9}));
+  const auto norm_r = builder.CreateVector<int8_t>(
+      reinterpret_cast<const signed char*>(norm_replacements),
+      sizeof(norm_replacements));
+  TrieBuilder trie_builder(builder);
+  trie_builder.add_nodes(trie_vector);
+  const auto norm_p = trie_builder.Finish();
+  EncoderConfigBuilder ecb(builder);
+  ecb.add_remove_extra_whitespaces(false);
+  ecb.add_normalized_prefixes(norm_p);
+  ecb.add_normalized_replacements(norm_r);
+  FinishEncoderConfigBuffer(builder, ecb.Finish());
+  const EncoderConfig* config = GetEncoderConfig(builder.GetBufferPointer());
+  {
+    const auto [res_string, offsets] =
+        NormalizeString("ABAABAAABAAAA", *config);
+    EXPECT_EQ(res_string, "A1BA2BA3BA4");
+    EXPECT_THAT(offsets,
+                ::testing::ElementsAre(0, 0, 1, 2, 2, 4, 5, 5, 8, 9, 9));
+  }
+}
+
+TEST(OptimizedEncoder, NormalizeStringWhitespacesRemove) {
+  flatbuffers::FlatBufferBuilder builder(1024);
+  const std::vector<std::string> norm_prefixes = {"A", "AA", "AAA", "AAAA",
+                                                  "X"};
+  const char norm_replacements[] = "A1\0A2\0A3\0A4\0 ";
+  const auto trie_vector =
+      builder.CreateVector(BuildTrie(norm_prefixes, {0, 3, 6, 9, 12}));
+  const auto norm_r = builder.CreateVector<int8_t>(
+      reinterpret_cast<const signed char*>(norm_replacements),
+      sizeof(norm_replacements));
+  TrieBuilder trie_builder(builder);
+  trie_builder.add_nodes(trie_vector);
+  const auto norm_p = trie_builder.Finish();
+  EncoderConfigBuilder ecb(builder);
+  ecb.add_remove_extra_whitespaces(true);
+  ecb.add_normalized_prefixes(norm_p);
+  ecb.add_normalized_replacements(norm_r);
+  FinishEncoderConfigBuffer(builder, ecb.Finish());
+  const EncoderConfig* config = GetEncoderConfig(builder.GetBufferPointer());
+  {
+    const auto [res_string, offsets] =
+        NormalizeString("XXABAABAAABAAAA", *config);
+    EXPECT_EQ(res_string, " A1BA2BA3BA4");
+    EXPECT_THAT(offsets,
+                ::testing::ElementsAre(0, 2, 2, 3, 4, 4, 6, 7, 7, 10, 11, 11));
+  }
+}
+
+TEST(OptimizedEncoder, ConfigConverter) {
+  std::string config;
+  auto status = internal::StdReadFileToString(kConfigFilePath, &config);
+  ASSERT_TRUE(status.ok());
+
+  ::sentencepiece::SentencePieceProcessor processor;
+  ASSERT_TRUE(processor.LoadFromSerializedProto(config).ok());
+  const auto converted_model = ConvertSentencepieceModel(config);
+  const std::string test_string("Hello world!\\xF0\\x9F\\x8D\\x95");
+  const auto encoded =
+      EncodeString(test_string, converted_model.data(), false, false, false);
+  ASSERT_EQ(encoded.codes.size(), encoded.offsets.size());
+
+  ::sentencepiece::SentencePieceText reference_encoded;
+  ASSERT_TRUE(processor.Encode(test_string, &reference_encoded).ok());
+  EXPECT_EQ(encoded.codes.size(), reference_encoded.pieces_size());
+  for (int i = 0; i < encoded.codes.size(); ++i) {
+    EXPECT_EQ(encoded.codes[i], reference_encoded.pieces(i).id());
+    EXPECT_EQ(encoded.offsets[i], reference_encoded.pieces(i).begin());
+  }
+}
+
+}  // namespace
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/py_tflite_registerer.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/py_tflite_registerer.cc
new file mode 100644
index 0000000..5345409
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/py_tflite_registerer.cc
@@ -0,0 +1,34 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/py_tflite_registerer.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+TfLiteRegistration* Register_SENTENCEPIECE_TOKENIZER();
+TfLiteRegistration* Register_SENTENCEPIECE_DETOKENIZER();
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+extern "C" void TFLite_SentencepieceTokenizerRegisterer(
+    tflite::MutableOpResolver* resolver) {
+  resolver->AddCustom("TFSentencepieceTokenizeOp",
+                      tflite::ops::custom::Register_SENTENCEPIECE_TOKENIZER());
+  resolver->AddCustom(
+      "TFSentencepieceDetokenizeOp",
+      tflite::ops::custom::Register_SENTENCEPIECE_DETOKENIZER());
+}
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/py_tflite_registerer.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/py_tflite_registerer.h
new file mode 100644
index 0000000..deb4e4e
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/py_tflite_registerer.h
@@ -0,0 +1,25 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_PY_TFLITE_REGISTERER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_PY_TFLITE_REGISTERER_H_
+#include "tensorflow/lite/mutable_op_resolver.h"
+
+// C-function that is called from the Python Wrapper.
+
+extern "C" void TFLite_SentencepieceTokenizerRegisterer(
+    tflite::MutableOpResolver *resolver);
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_PY_TFLITE_REGISTERER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_constants.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_constants.h
new file mode 100644
index 0000000..55644ba
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_constants.h
@@ -0,0 +1,43 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_CONSTANTS_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_CONSTANTS_H_
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+// The constant is copied from
+// https://github.com/google/sentencepiece/blob/master/src/unigram_model.cc
+constexpr float kUnkPenalty = 10.0;
+
+// These constants are copied from
+// https://github.com/google/sentencepiece/blob/master/src/sentencepiece_processor.cc
+//
+// Replaces white space with U+2581 (LOWER ONE EIGHT BLOCK).
+constexpr char kSpaceSymbol[] = "\xe2\x96\x81";
+
+// Encodes <unk> into U+2047 (DOUBLE QUESTION MARK),
+// since this character can be useful both for user and
+// developer. We can easily figure out that <unk> is emitted.
+constexpr char kDefaultUnknownSymbol[] = " \xE2\x81\x87 ";
+
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_CONSTANTS_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer.h
new file mode 100644
index 0000000..1f4e0f4
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer.h
@@ -0,0 +1,31 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_DETOKENIZER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_DETOKENIZER_H_
+
+// Constants are shared between TF and TFLite SentencepieceTokenizer kernels.
+namespace tensorflow {
+namespace ops {
+constexpr int kSPModelIndex = 0;
+constexpr int kInputIndex = 1;
+constexpr int kInputSplits = 2;
+constexpr int kAddBOSInput = 4;
+constexpr int kAddEOSInput = 5;
+constexpr int kReverseInput = 6;
+}  // namespace ops
+}  // namespace tensorflow
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_DETOKENIZER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer_op.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer_op.cc
new file mode 100644
index 0000000..bd4b5a1
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer_op.cc
@@ -0,0 +1,94 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow/core/framework/op.h"
+#include "tensorflow/core/framework/op_kernel.h"
+#include "tensorflow/core/framework/shape_inference.h"
+#include "tensorflow/core/framework/tensor.h"
+#include "tensorflow/core/protobuf/error_codes.pb.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer.h"
+
+namespace tensorflow {
+namespace ops {
+REGISTER_OP("TFSentencepieceDetokenizeOp")
+    .Input("sp_model: uint8")
+    .Input("input_values: int32")
+    .Input("input_splits: Tsplits")
+    .Attr("Tsplits: {int32, int64} = DT_INT64")
+    .Output("output: string")
+    .SetShapeFn([](tensorflow::shape_inference::InferenceContext* c) {
+      shape_inference::ShapeHandle unused;
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(1), 1, &unused));
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(2), 1, &unused));
+
+      shape_inference::DimensionHandle dim;
+      TF_RETURN_IF_ERROR(c->Subtract(c->NumElements(c->input(2)), 1, &dim));
+      c->set_output(0, c->Vector(dim));
+      return Status::OK();
+    });
+
+template <typename Tsplits>
+class TFSentencepieceDetokenizerOp : public tensorflow::OpKernel {
+ public:
+  explicit TFSentencepieceDetokenizerOp(tensorflow::OpKernelConstruction* ctx)
+      : OpKernel(ctx) {}
+  void Compute(tensorflow::OpKernelContext* ctx) override {
+    const auto& model_tensor = ctx->input(kSPModelIndex);
+    const auto& input_values_tensor = ctx->input(kInputIndex);
+    const auto input_values_flat =
+        input_values_tensor.flat<tensorflow::int32>();
+    const auto& input_splits_tensor = ctx->input(kInputSplits);
+    const auto input_splits_flat = input_splits_tensor.flat<Tsplits>();
+    const int num_of_sentences = input_splits_flat.size() - 1;
+    Tensor* output_tensor = nullptr;
+    OP_REQUIRES_OK(ctx,
+                   ctx->allocate_output(0, {num_of_sentences}, &output_tensor));
+    auto output_flat = output_tensor->flat<tensorflow::tstring>();
+    std::vector<int> codes_for_split;
+    int input_offset = 0;
+    for (int i = 0; i < num_of_sentences; i++) {
+      // Create a vector of int32 from input according to spans.
+      const int split_size = input_splits_flat(i + 1) - input_splits_flat(i);
+      codes_for_split.clear();
+      codes_for_split.reserve(split_size);
+      for (int j = 0; j < split_size; ++j) {
+        codes_for_split.push_back(input_values_flat(input_offset++));
+      }
+      const auto res = tflite::ops::custom::sentencepiece::DecodeString(
+          codes_for_split, model_tensor.data());
+      OP_REQUIRES(
+          ctx,
+          res.type ==
+              tflite::ops::custom::sentencepiece::DecoderResultType::SUCCESS,
+          tensorflow::Status(tensorflow::error::INTERNAL,
+                             "Sentencepiece conversion failed"));
+      output_flat(i) = res.decoded;
+    }
+  }
+};
+}  // namespace ops
+}  // namespace tensorflow
+
+REGISTER_KERNEL_BUILDER(
+    Name("TFSentencepieceDetokenizeOp")
+        .Device(tensorflow::DEVICE_CPU)
+        .TypeConstraint<tensorflow::int32>("Tsplits"),
+    tensorflow::ops::TFSentencepieceDetokenizerOp<tensorflow::int32>);
+REGISTER_KERNEL_BUILDER(
+    Name("TFSentencepieceDetokenizeOp")
+        .Device(tensorflow::DEVICE_CPU)
+        .TypeConstraint<tensorflow::int64>("Tsplits"),
+    tensorflow::ops::TFSentencepieceDetokenizerOp<tensorflow::int64>);
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer_tflite.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer_tflite.cc
new file mode 100644
index 0000000..54b34e4
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer_tflite.cc
@@ -0,0 +1,100 @@
+// Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+/**
+ * Sentencepiece tflite detokenizer implementation.
+ */
+#include <algorithm>
+#include <iterator>
+
+#include "flatbuffers/flexbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/context.h"
+#include "tensorflow/lite/kernels/internal/tensor.h"
+#include "tensorflow/lite/kernels/kernel_util.h"
+#include "tensorflow/lite/model.h"
+#include "tensorflow/lite/string_util.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_decoder.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_detokenizer.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+namespace detokenizer {
+
+constexpr int kOutputValuesInd = 0;
+// Initializes text encoder object from serialized parameters.
+void* Initialize(TfLiteContext* /*context*/, const char* /*buffer*/,
+                 size_t /*length*/) {
+  return nullptr;
+}
+void Free(TfLiteContext* /*context*/, void* /*buffer*/) {}
+
+TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
+  // TODO(mgubin): Add checks for input and output tensors.
+  TfLiteTensor& output_values =
+      context->tensors[node->outputs->data[kOutputValuesInd]];
+  SetTensorToDynamic(&output_values);
+  // TODO(mgubin): Check input types.
+
+  return kTfLiteOk;
+}
+
+TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
+  const TfLiteTensor& model_tensor =
+      context->tensors[node->inputs->data[tensorflow::ops::kSPModelIndex]];
+  const auto model_buffer_data = model_tensor.data.data;
+  const TfLiteTensor& input_encoded =
+      context->tensors[node->inputs->data[tensorflow::ops::kInputIndex]];
+  const int32_t* input_encoded_data = input_encoded.data.i32;
+  const TfLiteTensor& input_splits =
+      context->tensors[node->inputs->data[tensorflow::ops::kInputSplits]];
+  const int num_of_sentences = NumElements(input_splits.dims) - 1;
+  const int32_t* input_splits_data = input_splits.data.i32;
+
+  DynamicBuffer buf;
+
+  std::vector<int> codes_for_split;
+  int input_offset = 0;
+  for (int i = 0; i < num_of_sentences; i++) {
+    // Create a vector of int32 from input according to spans.
+    const int split_size = input_splits_data[i + 1] - input_splits_data[i];
+    codes_for_split.clear();
+    std::copy(input_encoded_data + input_offset,
+              input_encoded_data + input_offset + split_size,
+              std::back_inserter(codes_for_split));
+    const auto res = DecodeString(codes_for_split, model_buffer_data);
+    TF_LITE_ENSURE_MSG(context, res.type == DecoderResultType::SUCCESS,
+                       "Sentencepiece decoding failed");
+    buf.AddString(res.decoded.data(), res.decoded.length());
+    input_offset += split_size;
+  }
+  TfLiteTensor& output_values =
+      context->tensors[node->outputs->data[kOutputValuesInd]];
+  buf.WriteToTensor(&output_values, nullptr);
+  return kTfLiteOk;
+}
+}  // namespace detokenizer
+}  // namespace sentencepiece
+
+TfLiteRegistration* Register_SENTENCEPIECE_DETOKENIZER() {
+  static TfLiteRegistration r = {
+      sentencepiece::detokenizer::Initialize, sentencepiece::detokenizer::Free,
+      sentencepiece::detokenizer::Prepare, sentencepiece::detokenizer::Eval};
+  return &r;
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer.h
new file mode 100644
index 0000000..cb3ee07
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer.h
@@ -0,0 +1,31 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_TOKENIZER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_TOKENIZER_H_
+
+// Constants are shared between TF and TFLite SentencepieceTokenizer kernels.
+namespace tensorflow {
+namespace ops {
+
+constexpr int kSPModelIndex = 0;
+constexpr int kInputIndex = 1;
+constexpr int kAddBOSInput = 4;
+constexpr int kAddEOSInput = 5;
+constexpr int kReverseInput = 6;
+}  // namespace ops
+}  // namespace tensorflow
+
+#endif  // sTENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_SENTENCEPIECE_TOKENIZER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer_op.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer_op.cc
new file mode 100644
index 0000000..41fc5aa
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer_op.cc
@@ -0,0 +1,119 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <iterator>
+#include <vector>
+
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer.h"
+#include "tensorflow/core/framework/op.h"
+#include "tensorflow/core/framework/op_kernel.h"
+#include "tensorflow/core/framework/shape_inference.h"
+#include "tensorflow/core/framework/tensor.h"
+#include "tensorflow/core/protobuf/error_codes.pb.h"
+
+namespace tensorflow {
+namespace ops{
+
+// copied from third_party/tensorflow_text/core/ops/sentencepiece_ops.cc
+REGISTER_OP("TFSentencepieceTokenizeOp")
+    .Input("sp_model: uint8")
+    .Input("input: string")
+    .Input("nbest_size: int32")
+    .Input("alpha: float")
+    .Input("add_bos: bool")
+    .Input("add_eos: bool")
+    .Input("reverse: bool")
+    .Attr("out_type: {int32, string} = DT_INT32")
+    .Attr("Tsplits: {int32, int64} = DT_INT32")
+    .Output("output_values: out_type")
+    .Output("output_splits: Tsplits")
+    .SetShapeFn([](tensorflow::shape_inference::InferenceContext* c) {
+      tensorflow::shape_inference::ShapeHandle unused;
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(1), 1, &unused));
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(2), 0, &unused));
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(3), 0, &unused));
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(4), 0, &unused));
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(5), 0, &unused));
+      TF_RETURN_IF_ERROR(c->WithRank(c->input(6), 0, &unused));
+
+      c->set_output(
+          0, c->Vector(
+                 tensorflow::shape_inference::InferenceContext::kUnknownDim));
+
+      tensorflow::shape_inference::DimensionHandle num_splits;
+      TF_RETURN_IF_ERROR(c->Add(c->NumElements(c->input(1)), 1, &num_splits));
+      c->set_output(1, c->Vector(num_splits));
+      return tensorflow::Status::OK();
+    });
+
+class TFSentencepieceOp : public tensorflow::OpKernel {
+ public:
+  explicit TFSentencepieceOp(tensorflow::OpKernelConstruction* ctx)
+      : OpKernel(ctx) {}
+  void Compute(tensorflow::OpKernelContext* ctx) override {
+    const auto& model_tensor = ctx->input(kSPModelIndex);
+    const auto& input_values_tensor = ctx->input(kInputIndex);
+    const auto input_values_flat =
+        input_values_tensor.flat<tensorflow::tstring>();
+    const int num_of_input_values = input_values_flat.size();
+
+    const auto& add_bos_tensor = ctx->input(kAddBOSInput);
+    const bool add_bos = add_bos_tensor.scalar<bool>()();
+    const auto& add_eos_tensor = ctx->input(kAddEOSInput);
+    const bool add_eos = add_eos_tensor.scalar<bool>()();
+    const auto& reverse_tensor = ctx->input(kReverseInput);
+    const bool reverse = reverse_tensor.scalar<bool>()();
+
+    std::vector<int32> encoded;
+    std::vector<int32> splits;
+    for (int i = 0; i < num_of_input_values; ++i) {
+      const auto res = ::tflite::ops::custom::sentencepiece::EncodeString(
+          input_values_flat(i), model_tensor.data(), add_bos, add_eos, reverse);
+      OP_REQUIRES(
+          ctx,
+          res.type ==
+              ::tflite::ops::custom::sentencepiece::EncoderResultType::SUCCESS,
+          tensorflow::Status(tensorflow::error::INTERNAL,
+                             "Sentencepiece conversion failed"));
+      std::copy(res.codes.begin(), res.codes.end(),
+                std::back_inserter(encoded));
+      splits.emplace_back(encoded.size());
+    }
+    tensorflow::Tensor* output_values_tensor = nullptr;
+    tensorflow::Tensor* output_splits_tensor = nullptr;
+
+    OP_REQUIRES_OK(
+        ctx, ctx->allocate_output(0, {encoded.size()}, &output_values_tensor));
+    OP_REQUIRES_OK(ctx, ctx->allocate_output(1, {splits.size() + 1},
+                                             &output_splits_tensor));
+
+    auto values_tensor_flat = output_values_tensor->vec<int32>();
+    auto splits_tensor_flat = output_splits_tensor->vec<int32>();
+    for (int i = 0; i < encoded.size(); ++i) {
+      values_tensor_flat(i) = encoded[i];
+    }
+    splits_tensor_flat(0) = 0;
+    for (int i = 0; i < splits.size(); ++i) {
+      splits_tensor_flat(i + 1) = splits[i];
+    }
+  }
+};
+
+}  // namespace ops
+}  // namespace tensorflow
+REGISTER_KERNEL_BUILDER(
+    Name("TFSentencepieceTokenizeOp").Device(tensorflow::DEVICE_CPU),
+    tensorflow::ops::TFSentencepieceOp);
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer_tflite.cc b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer_tflite.cc
new file mode 100644
index 0000000..8309a6a
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer_tflite.cc
@@ -0,0 +1,129 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+/**
+ * Sentencepiece tflite tokenizer implementation.
+ */
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/optimized_encoder.h"
+#include "tensorflow_lite_support/custom_ops/kernel/sentencepiece/sentencepiece_tokenizer.h"
+#include "flatbuffers/flexbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/context.h"
+#include "tensorflow/lite/kernels/internal/tensor.h"
+#include "tensorflow/lite/kernels/kernel_util.h"
+#include "tensorflow/lite/model.h"
+#include "tensorflow/lite/string_util.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+namespace tokenizer {
+
+constexpr int kOutputValuesInd = 0;
+constexpr int kOutputSplitsInd = 1;
+
+namespace {
+TfLiteIntArray* CreateSizeArray(const std::initializer_list<int>& sizes) {
+  TfLiteIntArray* array_size = TfLiteIntArrayCreate(sizes.size());
+  int index = 0;
+  for (const int size : sizes) {
+    array_size->data[index++] = size;
+  }
+  return array_size;
+}
+}  // namespace
+
+// Initializes text encoder object from serialized parameters.
+void* Initialize(TfLiteContext* /*context*/, const char* /*buffer*/,
+                 size_t /*length*/) {
+  return nullptr;
+}
+void Free(TfLiteContext* /*context*/, void* /*buffer*/) {}
+
+TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
+  // TODO(mgubin): Add checks for input and output tensors.
+  TfLiteTensor& output_values =
+      context->tensors[node->outputs->data[kOutputValuesInd]];
+  SetTensorToDynamic(&output_values);
+
+  TfLiteTensor& output_splits =
+      context->tensors[node->outputs->data[kOutputSplitsInd]];
+  SetTensorToDynamic(&output_splits);
+  return kTfLiteOk;
+}
+
+TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
+  const TfLiteTensor& model_tensor =
+      context->tensors[node->inputs->data[tensorflow::ops::kSPModelIndex]];
+  const auto model_buffer_data = model_tensor.data.data;
+  const TfLiteTensor& input_text =
+      context->tensors[node->inputs->data[tensorflow::ops::kInputIndex]];
+
+  const TfLiteTensor add_bos_tensor =
+      context->tensors[node->inputs->data[tensorflow::ops::kAddBOSInput]];
+  const bool add_bos = add_bos_tensor.data.b[0];
+  const TfLiteTensor add_eos_tensor =
+      context->tensors[node->inputs->data[tensorflow::ops::kAddEOSInput]];
+  const bool add_eos = add_eos_tensor.data.b[0];
+  const TfLiteTensor reverse_tensor =
+      context->tensors[node->inputs->data[tensorflow::ops::kReverseInput]];
+  const bool reverse = reverse_tensor.data.b[0];
+
+  std::vector<int32> encoded;
+  std::vector<int32> splits;
+  const int num_strings = tflite::GetStringCount(&input_text);
+  for (int i = 0; i < num_strings; ++i) {
+    const auto strref = tflite::GetString(&input_text, i);
+    const auto res = EncodeString(std::string(strref.str, strref.len),
+                                  model_buffer_data, add_bos, add_eos, reverse);
+    TF_LITE_ENSURE_MSG(context, res.type == EncoderResultType::SUCCESS,
+                       "Sentencepiece conversion failed");
+    std::copy(res.codes.begin(), res.codes.end(), std::back_inserter(encoded));
+    splits.emplace_back(encoded.size());
+  }
+
+  TfLiteTensor& output_values =
+      context->tensors[node->outputs->data[kOutputValuesInd]];
+  TF_LITE_ENSURE_OK(context,
+                    context->ResizeTensor(
+                        context, &output_values,
+                        CreateSizeArray({static_cast<int>(encoded.size())})));
+  int32_t* output_values_flat = output_values.data.i32;
+  std::copy(encoded.begin(), encoded.end(), output_values_flat);
+  TfLiteTensor& output_splits =
+      context->tensors[node->outputs->data[kOutputSplitsInd]];
+  TF_LITE_ENSURE_OK(
+      context, context->ResizeTensor(
+                   context, &output_splits,
+                   CreateSizeArray({static_cast<int>(splits.size() + 1)})));
+  int32_t* output_splits_flat = output_splits.data.i32;
+  *output_splits_flat = 0;
+  std::copy(splits.begin(), splits.end(), output_splits_flat + 1);
+  return kTfLiteOk;
+}
+}  // namespace tokenizer
+}  // namespace sentencepiece
+
+TfLiteRegistration* Register_SENTENCEPIECE_TOKENIZER() {
+  static TfLiteRegistration r = {
+      sentencepiece::tokenizer::Initialize, sentencepiece::tokenizer::Free,
+      sentencepiece::tokenizer::Prepare, sentencepiece::tokenizer::Eval};
+  return &r;
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/testdata/sentencepiece.model b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/testdata/sentencepiece.model
new file mode 100644
index 0000000..041188f
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/testdata/sentencepiece.model
Binary files differ
diff --git a/tensorflow_lite_support/custom_ops/kernel/sentencepiece/utils.h b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/utils.h
new file mode 100644
index 0000000..13bc021
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/sentencepiece/utils.h
@@ -0,0 +1,66 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_UTILS_H_
+
+#include <ostream>
+#include <string>
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace sentencepiece {
+
+// AOSP and WASM doesn't support string_view,
+// we put here a minimal re-implementation.
+namespace utils {
+
+class string_view {
+ public:
+  explicit string_view(const std::string& s)
+      : str_(s.data()), len_(s.length()) {}
+  string_view(const char* str, int len) : str_(str), len_(len) {}
+  // A constructor from c string.
+  explicit string_view(const char* s) : str_(s), len_(strlen(s)) {}
+
+  int length() const { return len_; }
+  const char* data() const { return str_; }
+  bool empty() const { return len_ == 0; }
+  unsigned char at(int i) const { return str_[i]; }
+
+ private:
+  const char* str_ = nullptr;
+  const int len_ = 0;
+};
+
+inline std::ostream& operator<<(std::ostream& os, const string_view& sv) {
+  os << std::string(sv.data(), sv.length());
+  return os;
+}
+inline bool operator==(const string_view& view1, const string_view& view2) {
+  if (view1.length() != view2.length()) {
+    return false;
+  }
+  return memcmp(view1.data(), view2.data(), view1.length()) == 0;
+}
+
+}  // namespace utils
+}  // namespace sentencepiece
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_SENTENCEPIECE_UTILS_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_flex_delegate.tflite b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_flex_delegate.tflite
new file mode 100644
index 0000000..dc3b78b
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_flex_delegate.tflite
Binary files differ
diff --git a/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_ragged_1d_input.tflite b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_ragged_1d_input.tflite
new file mode 100644
index 0000000..03640e2
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_ragged_1d_input.tflite
Binary files differ
diff --git a/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_ragged_2d_input.tflite b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_ragged_2d_input.tflite
new file mode 100644
index 0000000..b688374
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_ragged_2d_input.tflite
Binary files differ
diff --git a/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_tensor.tflite b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_tensor.tflite
new file mode 100644
index 0000000..88e5cef
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/testdata/whitespace_tokenizer_to_tensor.tflite
Binary files differ
diff --git a/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.cc b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.cc
new file mode 100644
index 0000000..dad2f00
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.cc
@@ -0,0 +1,224 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.h"
+
+#include <algorithm>
+#include <utility>
+#include <vector>
+
+#include "tensorflow/lite/context.h"
+#include "tensorflow/lite/kernels/kernel_util.h"
+#include "tensorflow/lite/string_util.h"
+#include "libutf/utf.h"
+
+constexpr int kInput = 0;
+constexpr int kOutputValues = 0;
+constexpr int kOutputRowSplitsStart = 1;
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace whitespace_tokenizer {
+
+// This TFLite op implements a whitespace tokenizer, and can output the
+// tokens as either a padded tensor or a ragged tensor.
+//
+// If we're outputting a padded tensor, our outputs are:
+// * A string tensor
+//
+// If we're outputting a ragged tensor, our outputs are:
+// * A string tensor (the innermost values of the ragged tensor)
+// * N int64 tensors (the row_splits of the ragged tensor, where N is the
+//   rank of the input tensor)
+
+inline bool OutputIsPaddedTensor(TfLiteNode* node) {
+  return NumOutputs(node) == 1;
+}
+
+inline int charntorune(Rune* r, const char* s, int n) {
+  const int bytes_read = chartorune(r, const_cast<char *>(s));
+  if (bytes_read > n) {
+    *r = Runeerror;
+    return 0;
+  }
+  return bytes_read;
+}
+
+std::vector<std::pair<const char*, int>> Tokenize(StringRef str) {
+  const char* p = str.str;
+  int n = str.len;
+
+  std::vector<std::pair<const char*, int>> tokens;
+  const char* start = nullptr;
+  while (n > 0) {
+    Rune r;
+    int c = charntorune(&r, p, n);
+    if (r == Runeerror) break;
+
+    if (isspacerune(r)) {
+      if (start != nullptr) {
+        tokens.push_back({start, p - start});
+      }
+      start = nullptr;
+    } else {
+      if (start == nullptr) {
+        start = p;
+      }
+    }
+
+    p += c;
+    n -= c;
+  }
+  if (start != nullptr) {
+    tokens.push_back({start, p - start});
+  }
+
+  return tokens;
+}
+
+TfLiteStatus WritePaddedOutput(
+    const std::vector<std::vector<std::pair<const char*, int>>>& list_of_tokens,
+    const TfLiteTensor* input, TfLiteTensor* output_values) {
+  TfLiteIntArray* output_shape = TfLiteIntArrayCreate(NumDimensions(input) + 1);
+  for (int i = 0; i < NumDimensions(input); ++i) {
+    output_shape->data[i] = SizeOfDimension(input, i);
+  }
+
+  size_t max_tokens = 0;
+  for (const auto& tokens : list_of_tokens) {
+    max_tokens = std::max(max_tokens, tokens.size());
+  }
+
+  output_shape->data[NumDimensions(input)] = max_tokens;
+  DynamicBuffer buffer;
+  for (const auto& tokens : list_of_tokens) {
+    for (const auto& token : tokens) {
+      buffer.AddString(token.first, token.second);
+    }
+    for (int i = tokens.size(); i < max_tokens; ++i) {
+      buffer.AddString(nullptr, 0);
+    }
+  }
+  buffer.WriteToTensor(output_values, output_shape);
+  return kTfLiteOk;
+}
+
+TfLiteStatus WriteRaggedOutput(
+    const std::vector<std::vector<std::pair<const char*, int>>>& list_of_tokens,
+    const TfLiteTensor* input, TfLiteTensor* output_values,
+    std::vector<TfLiteTensor*> nested_row_splits) {
+  // The outer dimensions of the ragged tensor are all non-ragged.
+  for (int i = 0; i < nested_row_splits.size() - 1; ++i) {
+    int row_splits_step = SizeOfDimension(input, i + 1);
+    TfLiteTensor* row_splits = nested_row_splits[i];
+    for (int j = 0; j < SizeOfDimension(row_splits, 0); ++j) {
+      row_splits->data.i64[j] = j * row_splits_step;
+    }
+  }
+
+  // Generate the innermost row_splits and values tensors.
+  TfLiteTensor* row_splits = nested_row_splits.back();
+  TfLiteIntArray* output_shape = TfLiteIntArrayCreate(1);
+  DynamicBuffer buffer;
+  int token_index = 0;
+  int row_splits_index = 0;
+  for (const auto& tokens : list_of_tokens) {
+    row_splits->data.i64[row_splits_index] = token_index;
+    for (const auto& token : tokens) {
+      buffer.AddString(token.first, token.second);
+      ++token_index;
+    }
+    ++row_splits_index;
+  }
+  row_splits->data.i64[row_splits_index] = token_index;
+  output_shape->data[0] = token_index;
+  buffer.WriteToTensor(output_values, output_shape);
+  return kTfLiteOk;
+}
+
+TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
+  TfLiteTensor* output_values = GetOutput(context, node, kOutputValues);
+  SetTensorToDynamic(output_values);
+
+  if (OutputIsPaddedTensor(node)) {
+    return kTfLiteOk;
+  }
+
+  const TfLiteTensor* input = GetInput(context, node, kInput);
+  TF_LITE_ENSURE(context, NumDimensions(input) ==
+                              (NumOutputs(node) - kOutputRowSplitsStart));
+
+  // Resize the row_splits tensors.  We're just adding a ragged inner
+  // dimension to the shape of the input tensor, so the size of the
+  // row_splits tensors can be calculated using the input tensor's shape.
+  int input_size = 1;
+  for (int i = 0; i < NumDimensions(input); ++i) {
+    input_size *= SizeOfDimension(input, i);
+
+    TfLiteIntArray* row_splits_shape = TfLiteIntArrayCreate(1);
+    row_splits_shape->data[0] = input_size + 1;
+    TfLiteTensor* row_splits =
+        GetOutput(context, node, kOutputRowSplitsStart + i);
+    TF_LITE_ENSURE_STATUS(
+        context->ResizeTensor(context, row_splits, row_splits_shape));
+  }
+
+  return kTfLiteOk;
+}
+
+TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
+  const TfLiteTensor* input = GetInput(context, node, kInput);
+  int input_size = 1;
+  for (int i = 0; i < NumDimensions(input); ++i) {
+    input_size *= SizeOfDimension(input, i);
+  }
+
+  std::vector<std::vector<std::pair<const char*, int>>> list_of_tokens;
+  list_of_tokens.reserve(input_size);
+  for (int i = 0; i < input_size; ++i) {
+    list_of_tokens.emplace_back(Tokenize(GetString(input, i)));
+  }
+
+  TfLiteTensor* output_values = GetOutput(context, node, kOutputValues);
+  TF_LITE_ENSURE(context, IsDynamicTensor(output_values));
+
+  if (OutputIsPaddedTensor(node)) {
+    return WritePaddedOutput(list_of_tokens, input, output_values);
+  }
+
+  std::vector<TfLiteTensor*> nested_row_splits;
+  nested_row_splits.reserve(NumDimensions(input));
+  for (int i = 0; i < NumDimensions(input); ++i) {
+    TfLiteTensor* output_row_splits =
+        GetOutput(context, node, kOutputRowSplitsStart + i);
+    nested_row_splits.push_back(output_row_splits);
+  }
+  return WriteRaggedOutput(list_of_tokens, input, output_values,
+                           nested_row_splits);
+}
+
+}  // namespace whitespace_tokenizer
+
+TfLiteRegistration* Register_tftext_WhitespaceTokenizer() {
+  static TfLiteRegistration r = {nullptr, nullptr,
+                                 whitespace_tokenizer::Prepare,
+                                 whitespace_tokenizer::Eval};
+  return &r;
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.h b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.h
new file mode 100644
index 0000000..b190248
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.h
@@ -0,0 +1,31 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_WHITESPACE_TOKENIZER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_WHITESPACE_TOKENIZER_H_
+
+#include "tensorflow/lite/context.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+
+TfLiteRegistration* Register_tftext_WhitespaceTokenizer();
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_WHITESPACE_TOKENIZER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.cc b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.cc
new file mode 100644
index 0000000..534fbef
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.cc
@@ -0,0 +1,32 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.h"
+
+#include "tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.h"
+#include "tensorflow/lite/mutable_op_resolver.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+
+void AddWhitespaceTokenizerCustomOp(MutableOpResolver* resolver) {
+  resolver->AddCustom("tftext:WhitespaceTokenizer",
+                      Register_tftext_WhitespaceTokenizer());
+}
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.h b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.h
new file mode 100644
index 0000000..4f57d8d
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.h
@@ -0,0 +1,34 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_WHITESPACE_TOKENIZER_OP_RESOLVER_H_
+#define TENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_WHITESPACE_TOKENIZER_OP_RESOLVER_H_
+
+#include "tensorflow/lite/mutable_op_resolver.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+
+// Adds the WhitespaceTokenizer custom op to an op resolver.
+// This function can be loaded using dlopen.  Since C++ function names get
+// mangled, declare this function as extern C, so its name is unchanged.
+extern "C" void AddWhitespaceTokenizerCustomOp(MutableOpResolver* resolver);
+
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
+
+#endif  // LETENSORFLOW_LITE_SUPPORT_CUSTOM_OPS_KERNEL_WHITESPACE_TOKENIZER_OP_RESOLVER_H_
diff --git a/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver_wrapper.cc b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver_wrapper.cc
new file mode 100644
index 0000000..03d3ba8
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver_wrapper.cc
@@ -0,0 +1,29 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "pybind11/pybind11.h"
+#include "tensorflow/lite/mutable_op_resolver.h"
+#include "tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_op_resolver.h"
+
+PYBIND11_MODULE(_pywrap_whitespace_tokenizer_op_resolver, m) {
+  m.doc() = "_pywrap_whitespace_tokenizer_op_resolver";
+  m.def(
+      "AddWhitespaceTokenizerCustomOp",
+      [](uintptr_t resolver) {
+        tflite::ops::custom::AddWhitespaceTokenizerCustomOp(
+            reinterpret_cast<tflite::MutableOpResolver*>(resolver));
+      },
+      "Op registerer function for the tftext:WhitespaceTokenizer custom op.");
+}
diff --git a/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_test.cc b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_test.cc
new file mode 100644
index 0000000..4654e46
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_test.cc
@@ -0,0 +1,189 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer.h"
+
+#include <string>
+#include <vector>
+
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include "tensorflow/lite/kernels/test_util.h"
+#include "tensorflow/lite/schema/schema_generated.h"
+#include "tensorflow/lite/string_util.h"
+
+namespace tflite {
+namespace ops {
+namespace custom {
+namespace whitespace_tokenizer {
+namespace test {
+namespace {
+
+using ::testing::ElementsAre;
+using ::testing::ElementsAreArray;
+
+}  // namespace
+
+enum OutputType { PADDED, RAGGED };
+
+class WhitespaceTokenizerModel : public SingleOpModel {
+ public:
+  WhitespaceTokenizerModel(OutputType output_type,
+                           const std::vector<std::string>& input_values,
+                           const std::vector<int>& input_shape)
+      : input_shape_(input_shape) {
+    input_ = AddInput(TensorType_STRING);
+    output_values_ = AddOutput(TensorType_STRING);
+    if (output_type == RAGGED) {
+      for (int i = 0; i < input_shape_.size(); ++i) {
+        output_row_splits_.push_back(AddOutput(TensorType_INT64));
+      }
+    }
+    SetCustomOp("WhitespaceTokenizer", {}, Register_tftext_WhitespaceTokenizer);
+
+    BuildInterpreter({input_shape});
+    PopulateStringTensor(input_, input_values);
+    Invoke();
+  }
+
+  std::vector<int> GetValuesTensorShape() {
+    return GetTensorShape(output_values_);
+  }
+
+  std::vector<std::string> ExtractValuesTensorVector() {
+    std::vector<std::string> r;
+    TfLiteTensor* tensor = interpreter_->tensor(output_values_);
+    int n = GetStringCount(tensor);
+    for (int i = 0; i < n; ++i) {
+      StringRef ref = GetString(tensor, i);
+      r.emplace_back(ref.str, ref.len);
+    }
+    return r;
+  }
+
+  void CheckRowSplits(const std::vector<int>& token_counts) {
+    int size = 1;
+    for (int i = 0; i < input_shape_.size(); ++i) {
+      size *= input_shape_[i];
+      EXPECT_THAT(GetTensorShape(output_row_splits_[i]), ElementsAre(size + 1))
+          << "row_splits " << i << " has the wrong shape";
+
+      std::vector<int64_t> expected_values(size + 1);
+      if (i == input_shape_.size() - 1) {
+        ASSERT_EQ(token_counts.size(), size);
+
+        int index = 0;
+        expected_values[0] = index;
+        for (int j = 0; j < size; ++j) {
+          index += token_counts[j];
+          expected_values[j + 1] = index;
+        }
+      } else {
+        for (int j = 0; j <= size; ++j) {
+          expected_values[j] = j * input_shape_[i + 1];
+        }
+      }
+      EXPECT_THAT(ExtractVector<int64_t>(output_row_splits_[i]),
+                  ElementsAreArray(expected_values))
+          << "row_splits " << i << " has an incorrect value/index";
+    }
+  }
+
+ private:
+  int input_;
+  std::vector<int> input_shape_;
+  int output_values_;
+  std::vector<int> output_row_splits_;
+};  // namespace test
+
+TEST(WhitespaceTokenizerTest, SingleStringPaddedOutput) {
+  WhitespaceTokenizerModel m(PADDED, {"this is a test"}, {1});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(1, 4));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this", "is", "a", "test"));
+}
+
+TEST(WhitespaceTokenizerTest, SingleStringRaggedOutput) {
+  WhitespaceTokenizerModel m(RAGGED, {"this is a test"}, {1});
+  m.CheckRowSplits({4});
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("this", "is", "a", "test"));
+}
+
+TEST(WhitespaceTokenizerTest, VectorPaddedOutput) {
+  WhitespaceTokenizerModel m(PADDED,
+                             {"this is a test",        //
+                              "three token sentence",  //
+                              "many more tokens than that sentence"},
+                             {3});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(3, 6));
+  EXPECT_THAT(
+      m.ExtractValuesTensorVector(),
+      ElementsAre("this", "is", "a", "test", "", "",         //
+                  "three", "token", "sentence", "", "", "",  //
+                  "many", "more", "tokens", "than", "that", "sentence"));
+}
+
+TEST(WhitespaceTokenizerTest, VectorRaggedOutput) {
+  WhitespaceTokenizerModel m(RAGGED,
+                             {"this is a test",        //
+                              "three token sentence",  //
+                              "many more tokens than that sentence"},
+                             {3});
+  m.CheckRowSplits({4, 3, 6});
+  EXPECT_THAT(
+      m.ExtractValuesTensorVector(),
+      ElementsAre("this", "is", "a", "test",     //
+                  "three", "token", "sentence",  //
+                  "many", "more", "tokens", "than", "that", "sentence"));
+}
+
+TEST(WhitespaceTokenizerTest, MatrixPaddedOutput) {
+  WhitespaceTokenizerModel m(PADDED,
+                             {"a b c", "d e f",  //
+                              "g h", "i j k l",  //
+                              "m", "n o p q r"},
+                             {3, 2});
+  EXPECT_THAT(m.GetValuesTensorShape(), ElementsAre(3, 2, 5));
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("a", "b", "c", "", "",   //
+                          "d", "e", "f", "", "",   //
+                          "g", "h", "", "", "",    //
+                          "i", "j", "k", "l", "",  //
+                          "m", "", "", "", "",     //
+                          "n", "o", "p", "q", "r"));
+}
+
+TEST(WhitespaceTokenizerTest, MatrixRAGGEDOutput) {
+  WhitespaceTokenizerModel m(RAGGED,
+                             {"a b c", "d e f",  //
+                              "g h", "i j k l",  //
+                              "m", "n o p q r"},
+                             {3, 2});
+  m.CheckRowSplits({3, 3, 2, 4, 1, 5});
+  EXPECT_THAT(m.ExtractValuesTensorVector(),
+              ElementsAre("a", "b", "c",       //
+                          "d", "e", "f",       //
+                          "g", "h",            //
+                          "i", "j", "k", "l",  //
+                          "m",                 //
+                          "n", "o", "p", "q", "r"));
+}
+
+}  // namespace test
+}  // namespace whitespace_tokenizer
+}  // namespace custom
+}  // namespace ops
+}  // namespace tflite
diff --git a/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_test.py b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_test.py
new file mode 100644
index 0000000..b6a1a67
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/kernel/whitespace_tokenizer_test.py
@@ -0,0 +1,168 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+# Lint as: python3
+"""Tests for tensorflow_lite_support.custom_ops.kernel.whitespace_tokenizer."""
+
+import os
+import sys
+import timeit
+
+from absl import logging
+from absl.testing import parameterized
+import numpy as np
+import tensorflow as tf
+import tensorflow_text as tf_text
+# pylint: disable=g-direct-tensorflow-import
+from tensorflow.lite.python import interpreter as interpreter_wrapper
+from tensorflow.python.platform import resource_loader
+
+# Force loaded shared object symbols to be globally visible. This is needed so
+# that the interpreter_wrapper, in one .so file, can see the op resolver
+# in a different .so file. Note that this may already be set by default.
+# pylint: disable=g-import-not-at-top,g-bad-import-order,unused-import
+if hasattr(sys, 'setdlopenflags') and hasattr(sys, 'getdlopenflags'):
+  sys.setdlopenflags(sys.getdlopenflags() | os.RTLD_GLOBAL)
+from tensorflow_lite_support.custom_ops.kernel import _pywrap_whitespace_tokenizer_op_resolver
+
+TEST_CASES = [
+    ['this is a test'],
+    ['extra   spaces    in     here'],
+    ['a four token sentence', 'a five token sentence thing.'],
+    [['a multi dimensional test case', 'a b c d', 'e f g'],
+     ['h i j', 'k l m 2 3', 'n o p'], ['q r s 0 1', 't u v', 'w x y z']],
+]
+
+INVOKES_FOR_SINGLE_OP_BENCHMARK = 1000
+INVOKES_FOR_FLEX_DELEGATE_BENCHMARK = 10
+
+
+@tf.function
+def _call_whitespace_tokenizer_to_tensor(test_case):
+  tokenizer = tf_text.WhitespaceTokenizer()
+  return tokenizer.tokenize(test_case).to_tensor()
+
+
+@tf.function
+def _call_whitespace_tokenizer_to_ragged(test_case):
+  tokenizer = tf_text.WhitespaceTokenizer()
+  return tokenizer.tokenize(test_case)
+
+
+class WhitespaceTokenizerTest(parameterized.TestCase):
+
+  @parameterized.parameters([t] for t in TEST_CASES)
+  def testToTensorEquivalence(self, test_case):
+    tf_output = _call_whitespace_tokenizer_to_tensor(test_case)
+
+    model_filename = resource_loader.get_path_to_datafile(
+        'testdata/whitespace_tokenizer_to_tensor.tflite')
+    with open(model_filename, 'rb') as file:
+      model = file.read()
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=model,
+        custom_op_registerers=['AddWhitespaceTokenizerCustomOp'])
+
+    np_test_case = np.array(test_case, dtype=np.str)
+    interpreter.resize_tensor_input(0, np_test_case.shape)
+    interpreter.allocate_tensors()
+    interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                           np_test_case)
+    interpreter.invoke()
+    tflite_output = interpreter.get_tensor(
+        interpreter.get_output_details()[0]['index'])
+
+    self.assertEqual(tf_output.numpy().tolist(), tflite_output.tolist())
+
+  @parameterized.parameters([t] for t in TEST_CASES)
+  def testToRaggedEquivalence(self, test_case):
+    tf_output = _call_whitespace_tokenizer_to_ragged(test_case)
+
+    np_test_case = np.array(test_case, dtype=np.str)
+    rank = len(np_test_case.shape)
+
+    model_filename = resource_loader.get_path_to_datafile(
+        'testdata/whitespace_tokenizer_to_ragged_{}d_input.tflite'.format(rank))
+    with open(model_filename, 'rb') as file:
+      model = file.read()
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=model,
+        custom_op_registerers=['AddWhitespaceTokenizerCustomOp'])
+    interpreter.resize_tensor_input(0, np_test_case.shape)
+    interpreter.allocate_tensors()
+    interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                           np_test_case)
+    interpreter.invoke()
+
+    # Traverse the nested row_splits/values of the ragged tensor.
+    for i in range(rank):
+      tflite_output_cur_row_splits = interpreter.get_tensor(
+          interpreter.get_output_details()[1 + i]['index'])
+      self.assertEqual(tf_output.row_splits.numpy().tolist(),
+                       tflite_output_cur_row_splits.tolist())
+      tf_output = tf_output.values
+
+    tflite_output_values = interpreter.get_tensor(
+        interpreter.get_output_details()[0]['index'])
+    self.assertEqual(tf_output.numpy().tolist(), tflite_output_values.tolist())
+
+  def testSingleOpLatency(self):
+    model_filename = resource_loader.get_path_to_datafile(
+        'testdata/whitespace_tokenizer_to_tensor.tflite')
+    with open(model_filename, 'rb') as file:
+      model = file.read()
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=model,
+        custom_op_registerers=['AddWhitespaceTokenizerCustomOp'])
+
+    latency = 0.0
+    for test_case in TEST_CASES:
+      np_test_case = np.array(test_case, dtype=np.str)
+      interpreter.resize_tensor_input(0, np_test_case.shape)
+      interpreter.allocate_tensors()
+      interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                             np_test_case)
+      start_time = timeit.default_timer()
+      for _ in range(INVOKES_FOR_SINGLE_OP_BENCHMARK):
+        interpreter.invoke()
+      latency = latency + timeit.default_timer() - start_time
+
+    latency = latency / (INVOKES_FOR_SINGLE_OP_BENCHMARK * len(TEST_CASES))
+    logging.info('Latency: %fms', latency * 1000.0)
+
+  def testFlexDelegateLatency(self):
+    model_filename = resource_loader.get_path_to_datafile(
+        'testdata/whitespace_tokenizer_flex_delegate.tflite')
+    with open(model_filename, 'rb') as file:
+      model = file.read()
+    interpreter = interpreter_wrapper.Interpreter(model_content=model)
+
+    latency = 0.0
+    for test_case in TEST_CASES:
+      np_test_case = np.array(test_case, dtype=np.str)
+      interpreter.resize_tensor_input(0, np_test_case.shape)
+      interpreter.allocate_tensors()
+      interpreter.set_tensor(interpreter.get_input_details()[0]['index'],
+                             np_test_case)
+      start_time = timeit.default_timer()
+      for _ in range(INVOKES_FOR_FLEX_DELEGATE_BENCHMARK):
+        interpreter.invoke()
+      latency = latency + timeit.default_timer() - start_time
+
+    latency = latency / (INVOKES_FOR_FLEX_DELEGATE_BENCHMARK * len(TEST_CASES))
+    logging.info('Latency: %fms', latency * 1000.0)
+
+
+if __name__ == '__main__':
+  tf.test.main()
diff --git a/tensorflow_lite_support/custom_ops/python/BUILD b/tensorflow_lite_support/custom_ops/python/BUILD
new file mode 100644
index 0000000..82a8a6e
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/python/BUILD
@@ -0,0 +1,61 @@
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+py_library(
+    name = "tflite_text_api",
+    srcs = ["tflite_text_api.py"],
+    deps = [
+        # tensorflow dep,
+        # tensorflow_text dep,
+    ],
+)
+
+py_library(
+    name = "sentencepiece_tokenizer",
+    srcs = ["sentencepiece_tokenizer.py"],
+    data = [
+        "//tensorflow_lite_support/custom_ops/kernel/sentencepiece:sentencepiece_detokenizer_op.so",
+        "//tensorflow_lite_support/custom_ops/kernel/sentencepiece:sentencepiece_tokenizer_op.so",
+    ],
+    srcs_version = "PY3",
+    deps = [
+        # tensorflow dep,
+        "//tensorflow_lite_support/custom_ops/kernel/sentencepiece/py:pywrap_model_converter",
+    ],
+)
+
+py_test(
+    name = "sentencepiece_tokenizer_test",
+    srcs = ["sentencepiece_tokenizer_test.py"],
+    data = [
+        "//tensorflow_lite_support/custom_ops/kernel/sentencepiece:testdata",
+    ],
+    python_version = "PY3",
+    deps = [
+        ":sentencepiece_tokenizer",
+        # tensorflow dep,
+        # tensorflow_text dep,
+        "//tensorflow_lite_support/custom_ops/kernel/sentencepiece/py:pywrap_tflite_registerer",
+        "@absl_py//absl:app",
+        "@absl_py//absl/flags",
+        "@absl_py//absl/logging",
+        "@absl_py//absl/testing:parameterized",
+    ],
+)
+
+py_test(
+    name = "ragged_tensor_to_tensor_test",
+    srcs = ["ragged_tensor_to_tensor_test.py"],
+    python_version = "PY3",
+    deps = [
+        # tensorflow dep,
+        "//tensorflow_lite_support/custom_ops/kernel/ragged/py:pywrap_tflite_registerer",
+        "@absl_py//absl:app",
+        "@absl_py//absl/flags",
+        "@absl_py//absl/logging",
+    ],
+)
diff --git a/tensorflow_lite_support/custom_ops/python/ragged_tensor_to_tensor_test.py b/tensorflow_lite_support/custom_ops/python/ragged_tensor_to_tensor_test.py
new file mode 100644
index 0000000..319131e
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/python/ragged_tensor_to_tensor_test.py
@@ -0,0 +1,57 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# =============================================================================
+"""Tests for ragged_tensor_to_tensor."""
+
+import tensorflow as tf
+from tensorflow.lite.python import interpreter as interpreter_wrapper  # pylint: disable=g-direct-tensorflow-import
+
+
+class RaggedTensorToTensorTest(tf.test.TestCase):
+
+  def test_ragged_to_tensor(self):
+
+    @tf.function
+    def ragged_tensor_function():
+      ragged_tensor = tf.RaggedTensor.from_row_splits(
+          values=[
+              13, 36, 83, 131, 13, 36, 4, 3127, 152, 130, 30, 2424, 168, 1644,
+              1524, 4, 3127, 152, 130, 30, 2424, 168, 1644, 636
+          ],
+          row_splits=[0, 0, 6, 15, 24])
+      return ragged_tensor.to_tensor()
+
+    concrete_function = ragged_tensor_function.get_concrete_function()
+
+    converter = tf.lite.TFLiteConverter.from_concrete_functions(
+        [concrete_function])
+    converter.allow_custom_ops = True
+    tflite_model = converter.convert()
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=tflite_model,
+        custom_op_registerers=["TFLite_RaggedTensorToTensorRegisterer"])
+    interpreter.allocate_tensors()
+    interpreter.invoke()
+    output_details = interpreter.get_output_details()
+    expected_result_values = [[0, 0, 0, 0, 0, 0, 0, 0, 0],
+                              [13, 36, 83, 131, 13, 36, 0, 0, 0],
+                              [4, 3127, 152, 130, 30, 2424, 168, 1644, 1524],
+                              [4, 3127, 152, 130, 30, 2424, 168, 1644, 636]]
+    self.assertAllEqual(
+        interpreter.get_tensor(output_details[0]["index"]),
+        expected_result_values)
+
+
+if __name__ == "__main__":
+  tf.test.main()
diff --git a/tensorflow_lite_support/custom_ops/python/sentencepiece_tokenizer.py b/tensorflow_lite_support/custom_ops/python/sentencepiece_tokenizer.py
new file mode 100644
index 0000000..21efed5
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/python/sentencepiece_tokenizer.py
@@ -0,0 +1,125 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+# Lint as: python3
+"""Python class that implements Sentencepiece tokenizer.
+
+It follows TF.text designers design.
+
+"""
+import tensorflow.compat.v2 as tf  # pylint: disable=g-direct-tensorflow-import
+from tensorflow.python.ops.ragged import ragged_tensor  # pylint: disable=g-direct-tensorflow-import
+from tensorflow.python.framework import load_library
+from tensorflow.python.platform import resource_loader
+gen_sentencepiece_detokenizer_op = load_library.load_op_library(resource_loader.get_path_to_datafile('../kernel/sentencepiece/sentencepiece_detokenizer_op.so'))
+from tensorflow.python.framework import load_library
+from tensorflow.python.platform import resource_loader
+gen_sentencepiece_tokenizer_op = load_library.load_op_library(resource_loader.get_path_to_datafile('../kernel/sentencepiece/sentencepiece_tokenizer_op.so'))
+from tensorflow_lite_support.custom_ops.kernel.sentencepiece.py import pywrap_model_converter as model_converter
+
+
+class SentencepieceTokenizer:
+  """Sentencepiece tokenizer with tf.text interface."""
+
+  def __init__(self, model, reverse=False, add_bos=False, add_eos=False):
+    converted_model = model_converter.convert_sentencepiece_model(model)
+    converted_model_detokenizer = model_converter.convert_sentencepiece_model_for_decoder(
+        model)
+    # Use uint8 tensor as a buffer for the model to avoid any possible changes,
+    # for example truncation by '\0'.
+    self._converted_model = tf.constant(list(converted_model), dtype=tf.uint8)
+    self._converted_model_detokenizer = tf.constant(
+        list(converted_model_detokenizer), dtype=tf.uint8)
+    self._vocab_size = model_converter.get_vocabulary_size(converted_model)
+    self._reverse = reverse
+    self._add_bos = add_bos
+    self._add_eos = add_eos
+
+  def tokenize(self, inputs):
+    """The main tokenization function."""
+    input_tensor = ragged_tensor.convert_to_tensor_or_ragged_tensor(inputs)
+    if input_tensor.shape.ndims is None:
+      raise ValueError("Rank of input_tensor must be statically known.")
+    if ragged_tensor.is_ragged(input_tensor):
+      # Ensure that input has row_split_dtype is int32
+      input_tensor = input_tensor.with_row_splits_dtype(tf.int32)
+      # Recursively process the values of the ragged tensor.
+      tokens = self.tokenize(input_tensor.flat_values)
+      return input_tensor.with_flat_values(tokens)
+    else:
+      if input_tensor.shape.ndims > 1:
+        # Convert the input tensor to ragged and process it.
+        return self.tokenize(
+            tf.RaggedTensor.from_tensor(
+                input_tensor, row_splits_dtype=tf.int32))
+      elif input_tensor.shape.ndims == 0:
+        tokens = self.tokenize(tf.stack([input_tensor]))
+        return tokens.values
+      else:
+        # Our rank 1 tensor is the correct shape, so we can process it as
+        # normal.
+        (output_values, row_splits) = (
+            gen_sentencepiece_tokenizer_op.tf_sentencepiece_tokenize_op(
+                self._converted_model, input_tensor, 0, 0, self._add_bos,
+                self._add_eos, self._reverse))
+        tokens = tf.RaggedTensor.from_nested_row_splits(
+            flat_values=output_values,
+            nested_row_splits=[row_splits],
+            validate=False)
+        return tokens
+
+  def detokenize(self, input):  # pylint: disable=redefined-builtin
+    """Detokenizes tokens into preprocessed text.
+
+    Args:
+      input: A `RaggedTensor` or `Tensor` with int32 encoded text with rank >=
+        1.
+
+    Returns:
+      A N-1 dimensional string Tensor or RaggedTensor of the detokenized text.
+    """
+    input_tensor = ragged_tensor.convert_to_tensor_or_ragged_tensor(input)
+    if input_tensor.shape.ndims is None:
+      raise ValueError("Rank of input_tensor must be statically known.")
+    if input_tensor.shape.ndims == 0:
+      raise ValueError("Rank of input_tensor must be at least 1.")
+    if ragged_tensor.is_ragged(input_tensor):
+      if input_tensor.flat_values.shape.ndims > 1:
+        # If the flat_values of our ragged tensor is multi-dimensional, we can
+        # process it separately and our output will have the same nested
+        # splits as our input.
+        tokens = self.detokenize(input_tensor.flat_values)
+        return input_tensor.with_flat_values(tokens)
+      elif input_tensor.ragged_rank > 1:
+        # Recursively process the values of the ragged tensor.
+        tokens = self.detokenize(input_tensor.values)
+        return input_tensor.with_values(tokens)
+      else:
+        return gen_sentencepiece_detokenizer_op.tf_sentencepiece_detokenize_op(
+            self._converted_model_detokenizer, input_tensor.flat_values,
+            input_tensor.row_splits)
+    else:
+      if input_tensor.shape.ndims > 1:
+        # Convert the input tensor to ragged and process it.
+        return self.detokenize(
+            tf.RaggedTensor.from_tensor(
+                input_tensor, row_splits_dtype=tf.int32))
+      else:
+        tokens = self.detokenize(tf.stack([input_tensor]))
+        return tf.reshape(tokens, [])
+
+  def vocab_size(self):
+    """Returns size of the vocabulary in Sentencepiece model."""
+    return self._vocab_size
diff --git a/tensorflow_lite_support/custom_ops/python/sentencepiece_tokenizer_test.py b/tensorflow_lite_support/custom_ops/python/sentencepiece_tokenizer_test.py
new file mode 100644
index 0000000..3609b46
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/python/sentencepiece_tokenizer_test.py
@@ -0,0 +1,251 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+# Lint as: python3
+"""Tests for sentencepiece_tokenizer."""
+
+import os
+import sys
+import time
+
+from absl import flags
+import numpy as np
+import tensorflow.compat.v2 as tf  # pylint: disable=g-direct-tensorflow-import
+import tensorflow_text
+# Force loaded shared object symbols to be globally visible. This is needed so
+# that the interpreter_wrapper, in one .so file, can see the op resolver
+# in a different .so file. Note that this may already be set by default.
+# pylint: disable=g-import-not-at-top,g-bad-import-order,unused-import
+if hasattr(sys, "setdlopenflags") and hasattr(sys, "getdlopenflags"):
+  sys.setdlopenflags(sys.getdlopenflags() | os.RTLD_GLOBAL)
+from tensorflow.lite.python import interpreter as interpreter_wrapper  # pylint: disable=g-direct-tensorflow-import
+from tensorflow.python.platform import resource_loader
+from tensorflow_lite_support.custom_ops.python import sentencepiece_tokenizer
+from tensorflow_lite_support.custom_ops.kernel.sentencepiece.py import pywrap_tflite_registerer
+
+FLAGS = flags.FLAGS
+
+SENTENCEPIECE_MODEL_FILE = (
+    "../kernel/sentencepiece/testdata/sentencepiece.model")
+
+
+def _GetSentencepieceModel():
+  model_filename = resource_loader.get_path_to_datafile(
+      SENTENCEPIECE_MODEL_FILE)
+  with open(model_filename, "rb") as file:
+    model = file.read()
+  return model
+
+
+class SentencepieceTokenizerTest(tf.test.TestCase):
+
+  def setUp(self):
+    super(SentencepieceTokenizerTest, self).setUp()
+    self.sentencepiece_model = _GetSentencepieceModel()
+
+  def test_tftext_sentencepiece_tokenizer(self):
+    """Check that the new tokenizer produces the same result that the tftext one."""
+    tftext_sp = tensorflow_text.SentencepieceTokenizer(self.sentencepiece_model)
+    opt_sp = sentencepiece_tokenizer.SentencepieceTokenizer(
+        self.sentencepiece_model)
+
+    input_text = [
+        u" ", u"to be or not to be", u"ignored by length text1",
+        u"ignored by length text2"
+    ]
+    tftext_tokenized = tftext_sp.tokenize(input_text)
+    opt_tokenized = opt_sp.tokenize(input_text)
+    self.assertAllEqual(tftext_tokenized, opt_tokenized)
+
+  def test_tftext_sentencepiece_detokenizer(self):
+    """Check that the new tokenizer produces the same result that the tftext one."""
+    tftext_sp = tensorflow_text.SentencepieceTokenizer(self.sentencepiece_model)
+    opt_sp = sentencepiece_tokenizer.SentencepieceTokenizer(
+        self.sentencepiece_model)
+
+    input_text = [
+        u" ", u"to be or not to be", u"ignored by length text1",
+        u"ignored by length text2"
+    ]
+    tftext_tokenized = tftext_sp.tokenize(input_text)
+
+    # Check detokenizer
+    tftext_detokenized = tftext_sp.detokenize(tftext_tokenized)
+    opt_detokenized = opt_sp.detokenize(tftext_tokenized)
+    self.assertAllEqual(tftext_detokenized, opt_detokenized)
+
+  def test_tftext_sentencepiece_tokenizer_bos_eos(self):
+    """Check that the new tokenizer produces the same result that the tftext one with bos and eos."""
+    tftext_sp = tensorflow_text.SentencepieceTokenizer(
+        self.sentencepiece_model, add_bos=True, add_eos=True)
+    opt_sp = sentencepiece_tokenizer.SentencepieceTokenizer(
+        self.sentencepiece_model, add_bos=True, add_eos=True)
+
+    input_text = [
+        u" ", u"to be or not to be", u"ignored by length text1",
+        u"ignored by length text2"
+    ]
+    tftext_tokenized = tftext_sp.tokenize(input_text)
+    opt_tokenized = opt_sp.tokenize(input_text)
+    self.assertAllEqual(tftext_tokenized, opt_tokenized)
+
+  def test_tflite_opt_sentence_tokenizer(self):
+    """Check that can convert a Keras model to TFLite and it produces the same result for tokenization."""
+
+    class TokenizerLayer(tf.keras.layers.Layer):
+
+      def __init__(self, sentencepiece_model, **kwargs):
+        super(TokenizerLayer, self).__init__(**kwargs)
+        self.sp = sentencepiece_tokenizer.SentencepieceTokenizer(
+            sentencepiece_model)
+
+      def call(self, input_tensor, **kwargs):
+        return self.sp.tokenize(input_tensor).flat_values
+
+    model = tf.keras.models.Sequential(
+        [TokenizerLayer(self.sentencepiece_model)])
+    input_data = np.array([[
+        u" ", u"to be or not to be", u"ignored by length text1",
+        u"ignored by length text2"
+    ]])
+    tf_result = model.predict(input_data)
+    converter = tf.lite.TFLiteConverter.from_keras_model(model)
+    supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS]
+    converter.target_spec.supported_ops = supported_ops
+    converter.allow_custom_ops = True
+    tflite_model = converter.convert()
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=tflite_model,
+        custom_op_registerers=["TFLite_SentencepieceTokenizerRegisterer"])
+    interpreter.allocate_tensors()
+    input_details = interpreter.get_input_details()
+
+    interpreter.set_tensor(input_details[0]["index"], input_data)
+    interpreter.invoke()
+    output_details = interpreter.get_output_details()
+    expected_result = [
+        13, 36, 83, 131, 13, 36, 4, 3127, 152, 130, 30, 2424, 168, 1644, 1524,
+        4, 3127, 152, 130, 30, 2424, 168, 1644, 636
+    ]
+    self.assertAllEqual(tf_result, expected_result)
+    self.assertAllEqual(
+        interpreter.get_tensor(output_details[0]["index"]), expected_result)
+
+  def test_tflite_opt_sentence_detokenizer(self):
+    """Check that can convert a Keras model to TFLite and it produces the same result for tokenization."""
+
+    class DeTokenizerLayer(tf.keras.layers.Layer):
+
+      def __init__(self, sentencepiece_model, **kwargs):
+        super(DeTokenizerLayer, self).__init__(**kwargs)
+        self.sp = sentencepiece_tokenizer.SentencepieceTokenizer(
+            sentencepiece_model)
+
+      def call(self, input_tensor, **kwargs):
+        return self.sp.detokenize(input_tensor)
+
+    model = tf.keras.models.Sequential(
+        [DeTokenizerLayer(self.sentencepiece_model)])
+    input_data = np.array([[
+        13, 36, 83, 131, 13, 36, 4, 3127, 152, 130, 30, 2424, 168, 1644, 1524,
+        4, 3127, 152, 130, 30, 2424, 168, 1644, 636
+    ]],
+                          dtype=np.int32)
+    tf_result = model.predict(input_data)
+    converter = tf.lite.TFLiteConverter.from_keras_model(model)
+    supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS]
+    converter.target_spec.supported_ops = supported_ops
+    converter.allow_custom_ops = True
+    tflite_model = converter.convert()
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=tflite_model,
+        custom_op_registerers=["TFLite_SentencepieceTokenizerRegisterer"])
+    interpreter.allocate_tensors()
+    input_details = interpreter.get_input_details()
+
+    interpreter.set_tensor(input_details[0]["index"], input_data)
+    interpreter.invoke()
+    output_details = interpreter.get_output_details()
+    expected_result = [
+        "to be or not to be ignored by length text1 ignored by length text2"
+    ]
+    self.assertAllEqual(tf_result, expected_result)
+    self.assertAllEqual(
+        interpreter.get_tensor(output_details[0]["index"]), expected_result)
+
+  def test_tflite_opt_sentence_tokenizer_vocab_size(self):
+    """Check that can convert a Keras model to TFLite and it produces the same result for vocabulary size."""
+
+    class TokenizerLayer(tf.keras.layers.Layer):
+
+      def __init__(self, sentencepiece_model, **kwargs):
+        super(TokenizerLayer, self).__init__(**kwargs)
+        self.sp = sentencepiece_tokenizer.SentencepieceTokenizer(
+            sentencepiece_model)
+
+      def call(self, input_tensor, **kwargs):
+        return self.sp.vocab_size()
+
+    model = tf.keras.models.Sequential(
+        [TokenizerLayer(self.sentencepiece_model)])
+    input_data = np.array([[""]])
+    tf_result = model.predict(input_data)
+    converter = tf.lite.TFLiteConverter.from_keras_model(model)
+    supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS]
+    converter.target_spec.supported_ops = supported_ops
+    converter.allow_custom_ops = True
+    tflite_model = converter.convert()
+    interpreter = interpreter_wrapper.InterpreterWithCustomOps(
+        model_content=tflite_model,
+        custom_op_registerers=["TFLite_SentencepieceTokenizerRegisterer"])
+    interpreter.allocate_tensors()
+    input_details = interpreter.get_input_details()
+    interpreter.set_tensor(input_details[0]["index"], input_data)
+    interpreter.invoke()
+    output_details = interpreter.get_output_details()
+    expected_result = 4000
+    self.assertEqual(tf_result, expected_result)
+    self.assertAllEqual(
+        interpreter.get_tensor(output_details[0]["index"]), expected_result)
+
+
+class SentencepieceTokenizerBenchmark(tf.test.Benchmark):
+
+  def benchmarkTokenizer(self):
+    sp_model = _GetSentencepieceModel()
+    test_text = [
+        "This week we celebrate the casts and creatives who have come together"
+        " to bring us our favorite.",
+        "More Stacks products demonstrated commitment to excellent support.",
+        "Test, test, test."
+    ]
+
+    tftext_sp = tensorflow_text.SentencepieceTokenizer(sp_model)
+    opt_sp = sentencepiece_tokenizer.SentencepieceTokenizer(sp_model)
+    iter_number = 1000
+    start = time.time()
+    for _ in range(iter_number):
+      _ = opt_sp.tokenize(test_text)
+    self.report_benchmark(
+        iters=iter_number, wall_time=time.time() - start, name="opt")
+    start = time.time()
+    for _ in range(iter_number):
+      _ = tftext_sp.tokenize(test_text)
+    self.report_benchmark(
+        iters=iter_number, wall_time=time.time() - start, name="tf.text")
+
+
+if __name__ == "__main__":
+  tf.test.main()
diff --git a/tensorflow_lite_support/custom_ops/python/tflite_text_api.py b/tensorflow_lite_support/custom_ops/python/tflite_text_api.py
new file mode 100644
index 0000000..1466df2
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/python/tflite_text_api.py
@@ -0,0 +1,126 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+"""Wrapped TF.Text friendly to Tensorflow Lite conversion."""
+
+import tensorflow as tf
+import tensorflow_text as tf_text
+
+
+class WhitespaceTokenizer(tf_text.Tokenizer):
+  """TFLite friendly API for tensorflow_text.WhitspaceTokenizer.tokenize.
+
+  The strings are split on ICU defined whitespace characters. These
+  whitespace characters are dropped. See more details in
+  https://github.com/tensorflow/text/blob/master/docs/api_docs/python/text/WhitespaceTokenizer.md
+
+  Does not currently support tokenize_with_offsets().
+  """
+
+  def __init__(self):
+    super(WhitespaceTokenizer, self).__init__()
+    self._tokenizer = tf_text.WhitespaceTokenizer()
+
+  def tokenize(self, input_tensor):
+    """Tokenize input strings.
+
+    Args:
+      input_tensor: A `Tensor` of UTF-8 strings with rank 0, 1 or 2.
+
+    Returns:
+      A `RaggedTensor` of tokenized text. The returned shape is the shape of the
+      input tensor with an added ragged dimension for tokens of each string.
+    """
+
+    @tf.function(experimental_implements='name: "tftext:WhitespaceTokenizer"')
+    def func(input_tensor):
+      return self._tokenizer.tokenize(input_tensor)
+
+    return func(input_tensor)
+
+
+def ngrams(data,
+           width,
+           axis=-1,
+           reduction_type=None,
+           string_separator=' ',
+           name=None):
+  """TFLite friendly API for tensorflow_text.ngrams.
+
+  Creates a tensor of n-grams based data, a token tensor. See more details in
+  https://github.com/tensorflow/text/blob/master/docs/api_docs/python/text/ngrams.md
+
+  Args:
+    data: The data to reduce.  Must be convertible into a tf.Tensor or a
+      tf.RaggedTensor (in which case it will be deconstructed into its component
+      tf.Tensors).
+    width: The width of the ngram window. If there is not sufficient data to
+      fill out the ngram window, the resulting ngram will be empty.
+    axis: The axis to create ngrams along. Note that for string join reductions,
+      only axis '-1' is supported; for other reductions, any positive or
+      negative axis can be used. Should be a constant.
+    reduction_type: A member of the Reduction enum. Should be a constant.
+      Currently supports:
+      * `Reduction.STRING_JOIN`: Join strings in the window. Note that axis must
+        be -1 here.
+    string_separator: The separator string used for `Reduction.STRING_JOIN`.
+      Ignored otherwise. Must be a string constant, not a Tensor.
+    name: The op name.
+
+  Returns:
+    A tensor of ngrams.  If `data` is a ragged tensor, this will be a ragged
+    tensor.  Otherwise it will be a plain tensor.
+  """
+
+  if reduction_type is not tf_text.Reduction.STRING_JOIN:
+    # TODO(b/162082752): Provide support for Reduction.SUM and Reduction.MEAN
+    raise tf.errors.InvalidArgumentError(
+        None, None, 'only Reduction.STRING_JOIN is currently supported')
+
+  if reduction_type is tf_text.Reduction.STRING_JOIN and axis != -1:
+    raise tf.errors.InvalidArgumentError(
+        None, None, 'For Reduction.STRING_JOIN, axis must be -1')
+
+  experimental_implements = [
+      'name: "tftext:Ngrams"',
+      'attr { key: "width" value { i: %d } }' % width,
+      'attr { key: "axis" value { i: %d } }' % axis,
+      'attr { key: "reduction_type" value { s: "STRING_JOIN" } }',
+      'attr { key: "string_separator" value { s: "%s" } }' % string_separator,
+  ]
+  experimental_implements = ' '.join(experimental_implements)
+
+  if isinstance(data, tf.RaggedTensor):
+
+    # Since `data` can not be converted directly into a Tensor, we define
+    # ragged_func() which takes a deconstructed tf.RaggedTensor
+    # (one flat_values tensor and N row_splits tensors), pass it the
+    # deconstructed version of `data`, and then immediately reconstruct it
+    # within ragged_func().
+    @tf.function(experimental_implements=experimental_implements)
+    def ragged_func(values, *args):
+      ragged_tensor = tf.RaggedTensor.from_nested_row_splits(
+          flat_values=values, nested_row_splits=args)
+      return tf_text.ngrams(ragged_tensor, width, axis, reduction_type,
+                            string_separator, name)
+
+    return ragged_func(data.flat_values, *data.nested_row_splits)
+
+  @tf.function(experimental_implements=experimental_implements)
+  def func(data):
+    return tf_text.ngrams(data, width, axis, reduction_type, string_separator,
+                          name)
+
+  return func(data)
diff --git a/tensorflow_lite_support/custom_ops/testdata/sentencepiece_tokenizer_flex_op.tflite b/tensorflow_lite_support/custom_ops/testdata/sentencepiece_tokenizer_flex_op.tflite
new file mode 100644
index 0000000..e841b96
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/testdata/sentencepiece_tokenizer_flex_op.tflite
Binary files differ
diff --git a/tensorflow_lite_support/custom_ops/tf_configure.sh b/tensorflow_lite_support/custom_ops/tf_configure.sh
new file mode 100644
index 0000000..dbc96da
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/tf_configure.sh
@@ -0,0 +1,60 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+function write_action_env_to_bazelrc() {
+  echo "build --action_env $1=\"$2\"" >> .bazelrc
+}
+
+function is_linux() {
+  [[ "${PLATFORM}" == "linux" ]]
+}
+
+function is_macos() {
+  [[ "${PLATFORM}" == "darwin" ]]
+}
+
+function is_windows() {
+  # On windows, the shell script is actually running in msys
+  [[ "${PLATFORM}" =~ msys_nt*|mingw*|cygwin*|uwin* ]]
+}
+
+TF_CFLAGS=( $(python3 -c 'import tensorflow as tf; print(" ".join(tf.sysconfig.get_compile_flags()))') )
+TF_LFLAGS="$(python3 -c 'import tensorflow as tf; print(" ".join(tf.sysconfig.get_link_flags()))')"
+HEADER_DIR=${TF_CFLAGS:2}
+if is_windows; then
+  SHARED_LIBRARY_DIR=${SHARED_LIBRARY_DIR//\\//}
+  SHARED_LIBRARY_NAME=${SHARED_LIBRARY_NAME//\\//}
+  HEADER_DIR=${HEADER_DIR//\\//}
+fi
+if is_windows; then
+  # Use pywrap_tensorflow instead of tensorflow_framework on Windows
+  SHARED_LIBRARY_DIR=${TF_CFLAGS:2:-7}"python"
+else
+  SHARED_LIBRARY_DIR=${TF_LFLAGS:2}
+fi
+SHARED_LIBRARY_NAME=$(echo $TF_LFLAGS | rev | cut -d":" -f1 | rev)
+if ! [[ $TF_LFLAGS =~ .*:.* ]]; then
+  if is_macos; then
+    SHARED_LIBRARY_NAME="libtensorflow_framework.dylib"
+  elif is_windows; then
+    # Use pywrap_tensorflow's import library on Windows. It is in the same dir as the dll/pyd.
+    SHARED_LIBRARY_NAME="_pywrap_tensorflow_internal.lib"
+  else
+    SHARED_LIBRARY_NAME="libtensorflow_framework.so"
+  fi
+fi
+write_action_env_to_bazelrc "TF_HEADER_DIR" ${HEADER_DIR}
+write_action_env_to_bazelrc "TF_SHARED_LIBRARY_DIR" ${SHARED_LIBRARY_DIR}
+write_action_env_to_bazelrc "TF_SHARED_LIBRARY_NAME" ${SHARED_LIBRARY_NAME}
diff --git a/tensorflow_lite_support/custom_ops/tflite_inference_main.cc b/tensorflow_lite_support/custom_ops/tflite_inference_main.cc
new file mode 100644
index 0000000..2819dee
--- /dev/null
+++ b/tensorflow_lite_support/custom_ops/tflite_inference_main.cc
@@ -0,0 +1,105 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+// This program runs the tflite model specified in --model with random inputs.
+// For string type, the input is filled with a fixed string.
+
+#include <string>
+
+#include <glog/logging.h>
+#include "tensorflow/core/platform/init_main.h"
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/interpreter.h"
+#include "tensorflow/lite/kernels/register.h"
+#include "tensorflow/lite/model.h"
+#include "tensorflow/lite/model_builder.h"
+#include "tensorflow/lite/string_util.h"
+#include "tensorflow/lite/tools/command_line_flags.h"
+
+void FillRandomString(tflite::DynamicBuffer* buffer,
+                      const TfLiteIntArray* dim_array,
+                      const std::function<std::string()>& random_func) {
+  int num_elements = 1;
+  for (size_t i = 0; i < dim_array->size; i++) {
+    num_elements *= dim_array->data[i];
+  }
+  for (int i = 0; i < num_elements; ++i) {
+    auto str = random_func();
+    buffer->AddString(str.data(), str.length());
+  }
+}
+
+void RunWithRandomInputs(const std::string& filename) {
+  std::unique_ptr<tflite::FlatBufferModel> model =
+      tflite::FlatBufferModel::BuildFromFile(filename.c_str());
+
+  // Build the interpreter
+  tflite::ops::builtin::BuiltinOpResolver resolver;
+  std::unique_ptr<tflite::Interpreter> interpreter;
+  if (tflite::InterpreterBuilder(*model, resolver)(&interpreter) != kTfLiteOk) {
+    LOG(FATAL) << "Could not initialize interpreter for TFLite model.";
+  }
+
+  // Resize input tensors, if desired.
+  if (interpreter->AllocateTensors() != kTfLiteOk) {
+    LOG(FATAL) << "Could not allocate tensor.";
+  }
+
+  // Fill the random data.
+  std::vector<std::vector<uint8_t>> sample;
+  for (int tensor_idx : interpreter->inputs()) {
+    auto tensor = interpreter->tensor(tensor_idx);
+    if (tensor->type == kTfLiteString) {
+      tflite::DynamicBuffer buffer;
+      FillRandomString(&buffer, tensor->dims, []() {
+        return "we're have some friends over saturday to hang out in the "
+               "yard";
+      });
+      buffer.WriteToTensor(tensor, /*new_shape=*/nullptr);
+    } else {
+      std::vector<uint8_t> data(tensor->bytes);
+      for (auto it = data.begin(); it != data.end(); ++it) {
+        *it = random();
+      }
+      sample.push_back(data);
+      tensor->data.raw = reinterpret_cast<char*>(sample.rbegin()->data());
+    }
+  }
+
+  // Running inference.
+  if (interpreter->Invoke() != kTfLiteOk) {
+    LOG(FATAL) << "Failed to run the model.";
+  }
+
+  // Get the output.
+  for (int tensor_idx : interpreter->outputs()) {
+    auto tensor = interpreter->tensor(tensor_idx);
+    LOG(INFO) << "Output type: " << TfLiteTypeGetName(tensor->type);
+  }
+}
+
+int main(int argc, char** argv) {
+  // Parse flags to get the filename.
+  std::string filename;
+  std::vector<tflite::Flag> flag_list{tflite::Flag::CreateFlag(
+      "model", &filename, "The tflite model to run sample inference.",
+      tflite::Flag::kRequired)};
+  tflite::Flags::Parse(&argc, const_cast<const char**>(argv), flag_list);
+  tensorflow::port::InitMain(argv[0], &argc, &argv);
+
+  // Run the model with random inputs.
+  RunWithRandomInputs(filename);
+  return 0;
+}
diff --git a/tensorflow_lite_support/examples/task/text/desktop/BUILD b/tensorflow_lite_support/examples/task/text/desktop/BUILD
new file mode 100644
index 0000000..067d59c
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/text/desktop/BUILD
@@ -0,0 +1,68 @@
+package(
+    default_visibility = [
+        "//tensorflow_lite_support:users",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+# Example usage:
+# bazel run -c opt \
+#  tensorflow_lite_support/examples/task/text/desktop:bert_question_answerer_demo \
+#  -- \
+#  --model_path=/path/to/model.tflite \
+#  --question="question to ask" \
+#  --context="context for the question to ask"
+cc_binary(
+    name = "bert_question_answerer_demo",
+    srcs = ["bert_question_answerer_demo.cc"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/text/qa:bert_question_answerer",
+        "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/flags:parse",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
+
+# Example usage:
+# bazel run -c opt \
+#  tensorflow_lite_support/examples/task/text/desktop:bert_nl_classifier_demo \
+#  -- \
+#  --model_path=/path/to/model.tflite \
+#  --text="text to classify"
+cc_binary(
+    name = "bert_nl_classifier_demo",
+    srcs = ["bert_nl_classifier_demo.cc"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:category",
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:bert_nl_classifier",
+        "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/flags:parse",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
+
+# Example usage:
+# bazel run -c opt \
+#  tensorflow_lite_support/examples/task/text/desktop:nl_classifier_demo \
+#  -- \
+#  --model_path=/path/to/model.tflite \
+#  --text="text to classify" \
+#  --input_tensor_name="input_text" \
+#  --output_score_tensor_name="probability"
+cc_binary(
+    name = "nl_classifier_demo",
+    srcs = ["nl_classifier_demo.cc"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:category",
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier",
+        "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/flags:parse",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
diff --git a/tensorflow_lite_support/examples/task/text/desktop/README.md b/tensorflow_lite_support/examples/task/text/desktop/README.md
new file mode 100644
index 0000000..859504f
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/text/desktop/README.md
@@ -0,0 +1,134 @@
+# CLI Demos for C++ Text Task APIs
+
+This folder contains simple command-line tools for easily trying out the C++
+Text Task APIs.
+
+## Bert Question Answerer
+
+#### Prerequisites
+
+You will need:
+
+* a TFLite bert based question answerer model from model maker.
+(e.g. [mobilebert][1] or [albert][2] available on TensorFlow Hub).
+
+#### Usage
+
+In the console, run:
+
+```bash
+# Download the model:
+curl \
+ -L 'https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1?lite-format=tflite' \
+ -o /tmp/mobilebert.tflite
+
+# Run the classification tool:
+bazel run -c opt \
+ tensorflow_lite_support/examples/task/text/desktop:bert_question_answerer_demo -- \
+ --model_path=/tmp/mobilebert.tflite \
+ --question="Where is Amazon rainforest?" \
+ --context="The Amazon rainforest, alternatively, the Amazon Jungle, also known in \
+English as Amazonia, is a moist broadleaf tropical rainforest in the Amazon \
+biome that covers most of the Amazon basin of South America. This basin \
+encompasses 7,000,000 km2 (2,700,000 sq mi), of which \
+5,500,000 km2 (2,100,000 sq mi) are covered by the rainforest. This region \
+includes territory belonging to nine nations."
+```
+
+#### Results
+
+In the console, you should get:
+
+```
+answer[0]:  'South America.'
+logit: 1.84847, start_index: 39, end_index: 40
+answer[1]:  'most of the Amazon basin of South America.'
+logit: 1.2921, start_index: 34, end_index: 40
+answer[2]:  'the Amazon basin of South America.'
+logit: -0.0959535, start_index: 36, end_index: 40
+answer[3]:  'the Amazon biome that covers most of the Amazon basin of South America.'
+logit: -0.498558, start_index: 28, end_index: 40
+answer[4]:  'Amazon basin of South America.'
+logit: -0.774266, start_index: 37, end_index: 40
+```
+
+## NLClassifier
+
+#### Prerequisites
+
+You will need:
+
+* a TFLite text classification model with certain format.
+(e.g. [movie_review_model][3], a model to classify movie reviews), you'll need
+to configure the input tensor and out tensor for the API, see the [doc][4] for 
+details.
+
+#### Usage
+
+In the console, run:
+
+```bash
+# Download the model:
+curl \
+ -L 'https://storage.googleapis.com/download.tensorflow.org/models/tflite/text_classification/text_classification_v2.tflite' \
+ -o /tmp/movie_review.tflite
+
+# Run the detection tool:
+bazel run -c opt \
+ tensorflow_lite_support/examples/task/text/desktop:nl_classifier_demo -- \
+ --model_path=/tmp/movie_review.tflite \
+ --text="What a waste of my time." \
+ --input_tensor_name="input_text" \
+ --output_score_tensor_name="probability"
+```
+
+#### Results
+
+In the console, you should get:
+
+```
+category[0]: 'Negative' : '0.81313'
+category[1]: 'Positive' : '0.18687'
+```
+
+## BertNLClassifier
+
+#### Prerequisites
+
+TODO(b/163086702): Update the links to models with metadata attached.
+
+You will need:
+
+* a Bert based TFLite text classification model from model maker. (e.g. [movie_review_model][5] available on TensorFlow Hub).
+
+#### Usage
+
+In the console, run:
+
+```bash
+# Download the model:
+curl \
+ -L 'https://url/to/bert/nl/classifier' \
+ -o /tmp/bert_movie_review.tflite
+
+# Run the segmentation tool:
+bazel run -c opt \
+ tensorflow_lite_support/examples/task/text/desktop:bert_nl_classifier_demo -- \
+ --model_path=/tmp/bert_movie_review.tflite \
+ --text="it's a charming and often affecting journey"
+```
+
+#### Results
+
+In the console, you should get:
+
+```
+category[0]: 'negative' : '0.00006'
+category[1]: 'positive' : '0.99994'
+```
+
+[1]: https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1
+[2]: https://tfhub.dev/tensorflow/lite-model/albert_lite_base/squadv1/1
+[3]: https://www.tensorflow.org/lite/models/text_classification/overview
+[4]: https://github.com/tensorflow/tflite-support/blob/fe8b69002f5416900285dc69e2baa078c91bd994/tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h#L55
+[5]: http://bert/nl/classifier/model
diff --git a/tensorflow_lite_support/examples/task/text/desktop/bert_nl_classifier_demo.cc b/tensorflow_lite_support/examples/task/text/desktop/bert_nl_classifier_demo.cc
new file mode 100644
index 0000000..15ea3bf
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/text/desktop/bert_nl_classifier_demo.cc
@@ -0,0 +1,77 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include <iostream>
+#include <limits>
+
+#include "absl/flags/flag.h"
+#include "absl/flags/parse.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.h"
+
+ABSL_FLAG(std::string, model_path, "",
+          "Absolute path to the '.tflite' bert classification model.");
+ABSL_FLAG(std::string, text, "", "Text to classify.");
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+absl::Status Classify() {
+  ASSIGN_OR_RETURN(
+      std::unique_ptr<BertNLClassifier> classifier,
+      BertNLClassifier::CreateFromFile(absl::GetFlag(FLAGS_model_path)));
+
+  std::vector<core::Category> categories =
+      classifier->Classify(absl::GetFlag(FLAGS_text));
+
+  for (int i = 0; i < categories.size(); ++i) {
+    const core::Category& category = categories[i];
+    std::cout << absl::StrFormat("category[%d]: '%s' : '%.5f'\n", i,
+                                 category.class_name, category.score);
+  }
+
+  return absl::OkStatus();
+}
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+int main(int argc, char** argv) {
+  // Parse command line arguments and perform sanity checks.
+  absl::ParseCommandLine(argc, argv);
+  if (absl::GetFlag(FLAGS_model_path).empty()) {
+    std::cerr << "Missing mandatory 'model_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_text).empty()) {
+    std::cerr << "Missing mandatory 'question' argument.\n";
+    return 1;
+  }
+
+  // Run classification.
+  absl::Status status = tflite::task::text::nlclassifier::Classify();
+  if (status.ok()) {
+    return 0;
+  } else {
+    std::cerr << "Classification failed: " << status.message() << "\n";
+    return 1;
+  }
+}
diff --git a/tensorflow_lite_support/examples/task/text/desktop/bert_question_answerer_demo.cc b/tensorflow_lite_support/examples/task/text/desktop/bert_question_answerer_demo.cc
new file mode 100644
index 0000000..743db71
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/text/desktop/bert_question_answerer_demo.cc
@@ -0,0 +1,81 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include <iostream>
+#include <limits>
+
+#include "absl/flags/flag.h"
+#include "absl/flags/parse.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.h"
+
+ABSL_FLAG(std::string, model_path, "",
+          "Absolute path to the '.tflite' bert question answerer model.");
+ABSL_FLAG(std::string, question, "", "Question to ask.");
+ABSL_FLAG(std::string, context, "",
+          "Context the asked question is based upon.");
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace qa {
+
+absl::Status Answer() {
+  ASSIGN_OR_RETURN(
+      std::unique_ptr<QuestionAnswerer> answerer,
+      BertQuestionAnswerer::CreateFromFile(absl::GetFlag(FLAGS_model_path)));
+
+  std::vector<QaAnswer> answers = answerer->Answer(
+      absl::GetFlag(FLAGS_context), absl::GetFlag(FLAGS_question));
+  for (int i = 0; i < answers.size(); ++i) {
+    const QaAnswer& answer = answers[i];
+    std::cout << absl::StrFormat(
+        "answer[%d]: '%s'\n    logit: '%.5f, start_index: %d, end_index: %d\n",
+        i, answer.text, answer.pos.logit, answer.pos.start, answer.pos.end);
+  }
+
+  return absl::OkStatus();
+}
+
+}  // namespace qa
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+int main(int argc, char** argv) {
+  // Parse command line arguments and perform sanity checks.
+  absl::ParseCommandLine(argc, argv);
+  if (absl::GetFlag(FLAGS_model_path).empty()) {
+    std::cerr << "Missing mandatory 'model_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_question).empty()) {
+    std::cerr << "Missing mandatory 'question' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_context).empty()) {
+    std::cerr << "Missing mandatory 'context' argument.\n";
+    return 1;
+  }
+  // Run the answerer.
+  absl::Status status = tflite::task::text::qa::Answer();
+  if (status.ok()) {
+    return 0;
+  } else {
+    std::cerr << "Answer failed: " << status.message() << "\n";
+    return 1;
+  }
+}
diff --git a/tensorflow_lite_support/examples/task/text/desktop/nl_classifier_demo.cc b/tensorflow_lite_support/examples/task/text/desktop/nl_classifier_demo.cc
new file mode 100644
index 0000000..2e96ec6
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/text/desktop/nl_classifier_demo.cc
@@ -0,0 +1,112 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include <iostream>
+#include <limits>
+
+#include "absl/flags/flag.h"
+#include "absl/flags/parse.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/category.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h"
+
+ABSL_FLAG(std::string, model_path, "",
+          "Absolute path to the '.tflite' classification model.");
+ABSL_FLAG(std::string, text, "", "Text to classify.");
+ABSL_FLAG(int, input_tensor_index, -1, "Input tensor index of the model.");
+ABSL_FLAG(int, output_score_tensor_index, -1,
+          "Output score tensor index of the model.");
+ABSL_FLAG(int, output_label_tensor_index, -1,
+          "Output label tensor index of the model.");
+ABSL_FLAG(std::string, input_tensor_name, "",
+          "Input tensor name of the model.");
+ABSL_FLAG(std::string, output_score_tensor_name, "",
+          "Output score tensor name of the model.");
+ABSL_FLAG(std::string, output_label_tensor_name, "",
+          "Output label tensor name of the model.");
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+absl::Status Classify() {
+  NLClassifierOptions options{};
+  if (absl::GetFlag(FLAGS_input_tensor_index) >= 0) {
+    options.input_tensor_index = absl::GetFlag(FLAGS_input_tensor_index);
+  }
+  if (absl::GetFlag(FLAGS_output_score_tensor_index) >= 0) {
+    options.output_score_tensor_index =
+        absl::GetFlag(FLAGS_output_score_tensor_index);
+  }
+  if (absl::GetFlag(FLAGS_output_label_tensor_index) >= 0) {
+    options.output_label_tensor_index =
+        absl::GetFlag(FLAGS_output_label_tensor_index);
+  }
+  if (!absl::GetFlag(FLAGS_input_tensor_name).empty()) {
+    options.input_tensor_name = absl::GetFlag(FLAGS_input_tensor_name);
+  }
+  if (!absl::GetFlag(FLAGS_output_score_tensor_name).empty()) {
+    options.output_score_tensor_name =
+        absl::GetFlag(FLAGS_output_score_tensor_name);
+  }
+  if (!absl::GetFlag(FLAGS_output_label_tensor_name).empty()) {
+    options.output_label_tensor_name =
+        absl::GetFlag(FLAGS_output_label_tensor_name);
+  }
+
+  ASSIGN_OR_RETURN(std::unique_ptr<NLClassifier> classifier,
+                   NLClassifier::CreateFromFileAndOptions(
+                       absl::GetFlag(FLAGS_model_path), options));
+
+  std::vector<core::Category> categories =
+      classifier->Classify(absl::GetFlag(FLAGS_text));
+
+  for (int i = 0; i < categories.size(); ++i) {
+    const core::Category& category = categories[i];
+    std::cout << absl::StrFormat("category[%d]: '%s' : '%.5f'\n", i,
+                                 category.class_name, category.score);
+  }
+
+  return absl::OkStatus();
+}
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+int main(int argc, char** argv) {
+  // Parse command line arguments and perform sanity checks.
+  absl::ParseCommandLine(argc, argv);
+  if (absl::GetFlag(FLAGS_model_path).empty()) {
+    std::cerr << "Missing mandatory 'model_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_text).empty()) {
+    std::cerr << "Missing mandatory 'question' argument.\n";
+    return 1;
+  }
+
+  // Run classification.
+  absl::Status status = tflite::task::text::nlclassifier::Classify();
+  if (status.ok()) {
+    return 0;
+  } else {
+    std::cerr << "Classification failed: " << status.message() << "\n";
+    return 1;
+  }
+}
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/BUILD b/tensorflow_lite_support/examples/task/vision/desktop/BUILD
new file mode 100644
index 0000000..f61984e
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/BUILD
@@ -0,0 +1,68 @@
+package(
+    default_visibility = [
+        "//tensorflow_lite_support:users",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_binary(
+    name = "image_classifier_demo",
+    srcs = ["image_classifier_demo.cc"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:external_file_handler",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision:image_classifier",
+        "//tensorflow_lite_support/cc/task/vision/proto:class_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:classifications_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_classifier_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
+        "//tensorflow_lite_support/examples/task/vision/desktop/utils:image_utils",
+        "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/flags:parse",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
+
+cc_binary(
+    name = "object_detector_demo",
+    srcs = ["object_detector_demo.cc"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:external_file_handler",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision:object_detector",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:class_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:detections_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:object_detector_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
+        "//tensorflow_lite_support/examples/task/vision/desktop/utils:image_utils",
+        "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/flags:parse",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
+
+cc_binary(
+    name = "image_segmenter_demo",
+    srcs = ["image_segmenter_demo.cc"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/core:external_file_handler",
+        "//tensorflow_lite_support/cc/task/core/proto:external_file_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision:image_segmenter",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_segmenter_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:segmentations_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
+        "//tensorflow_lite_support/examples/task/vision/desktop/utils:image_utils",
+        "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/flags:parse",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+    ],
+)
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/README.md b/tensorflow_lite_support/examples/task/vision/desktop/README.md
new file mode 100644
index 0000000..73c6b63
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/README.md
@@ -0,0 +1,180 @@
+# CLI Demos for C++ Vision Task APIs
+
+This folder contains simple command-line tools for easily trying out the C++
+Vision Task APIs.
+
+## Image Classifier
+
+#### Prerequisites
+
+You will need:
+
+* a TFLite image classification model (e.g. [aiy/vision/classifier/birds_V1][1],
+a bird classification model available on TensorFlow Hub),
+* a PNG, JPEG or GIF image to run classification on, e.g.:
+
+![sparrow](g3doc/sparrow.jpg)
+
+#### Usage
+
+In the console, run:
+
+```bash
+# Download the model:
+curl \
+ -L 'https://tfhub.dev/google/lite-model/aiy/vision/classifier/birds_V1/3?lite-format=tflite' \
+ -o /tmp/aiy_vision_classifier_birds_V1_3.tflite
+
+# Run the classification tool:
+bazel run -c opt \
+ tensorflow_lite_support/examples/task/vision/desktop:image_classifier_demo -- \
+ --model_path=/tmp/aiy_vision_classifier_birds_V1_3.tflite \
+ --image_path=\
+$(pwd)/tensorflow_lite_support/examples/task/vision/desktop/g3doc/sparrow.jpg \
+ --max_results=3
+```
+
+#### Results
+
+In the console, you should get:
+
+```
+Results:
+  Rank #0:
+   index       : 671
+   score       : 0.91406
+   class name  : /m/01bwb9
+   display name: Passer domesticus
+  Rank #1:
+   index       : 670
+   score       : 0.00391
+   class name  : /m/01bwbt
+   display name: Passer montanus
+  Rank #2:
+   index       : 495
+   score       : 0.00391
+   class name  : /m/0bwm6m
+   display name: Passer italiae
+```
+
+## Object Detector
+
+#### Prerequisites
+
+You will need:
+
+* a TFLite object detection model (e.g. [ssd_mobilenet_v1][2], a generic object
+detection model available on TensorFlow Hub),
+* a PNG, JPEG or GIF image to run detection on, e.g.:
+
+![dogs](g3doc/dogs.jpg)
+
+#### Usage
+
+In the console, run:
+
+```bash
+# Download the model:
+curl \
+ -L 'https://tfhub.dev/tensorflow/lite-model/ssd_mobilenet_v1/1/metadata/1?lite-format=tflite' \
+ -o /tmp/ssd_mobilenet_v1_1_metadata_1.tflite
+
+# Run the detection tool:
+bazel run -c opt \
+ tensorflow_lite_support/examples/task/vision/desktop:object_detector_demo -- \
+ --model_path=/tmp/ssd_mobilenet_v1_1_metadata_1.tflite \
+ --image_path=\
+$(pwd)/tensorflow_lite_support/examples/task/vision/desktop/g3doc/dogs.jpg \
+ --output_png=/tmp/detection-output.png \
+ --max_results=2
+```
+
+#### Results
+
+In the console, you should get:
+
+```
+Results saved to: /tmp/detection-output.png
+Results:
+ Detection #0 (red):
+  Box: (x: 355, y: 133, w: 190, h: 206)
+  Top-1 class:
+   index       : 17
+   score       : 0.73828
+   class name  : dog
+ Detection #1 (green):
+  Box: (x: 103, y: 15, w: 138, h: 369)
+  Top-1 class:
+   index       : 17
+   score       : 0.73047
+   class name  : dog
+```
+
+And `/tmp/detection-output.jpg` should contain:
+
+![detection-output](g3doc/detection-output.png)
+
+## Image Segmenter
+
+#### Prerequisites
+
+You will need:
+
+* a TFLite image segmentation model (e.g. [deeplab_v3][3], a generic
+segmentation model available on TensorFlow Hub),
+* a PNG, JPEG or GIF image to run segmentation on, e.g.:
+
+![plane](g3doc/plane.jpg)
+
+#### Usage
+
+In the console, run:
+
+```bash
+# Download the model:
+curl \
+ -L 'https://tfhub.dev/tensorflow/lite-model/deeplabv3/1/metadata/1?lite-format=tflite' \
+ -o /tmp/deeplabv3_1_metadata_1.tflite
+
+# Run the segmentation tool:
+bazel run -c opt \
+ tensorflow_lite_support/examples/task/vision/desktop:image_segmenter_demo -- \
+ --model_path=/tmp/deeplabv3_1_metadata_1.tflite \
+ --image_path=\
+$(pwd)/tensorflow_lite_support/examples/task/vision/desktop/g3doc/plane.jpg \
+ --output_mask_png=/tmp/segmentation-output.png
+```
+
+#### Results
+
+In the console, you should get:
+
+```
+Category mask saved to: /tmp/segmentation-output.png
+Color Legend:
+ (r: 000, g: 000, b: 000):
+  index       : 0
+  class name  : background
+ (r: 128, g: 000, b: 000):
+  index       : 1
+  class name  : aeroplane
+
+# (omitting multiple lines for conciseness) ...
+
+ (r: 128, g: 192, b: 000):
+  index       : 19
+  class name  : train
+ (r: 000, g: 064, b: 128):
+  index       : 20
+  class name  : tv
+Tip: use a color picker on the output PNG file to inspect the output mask with
+this legend.
+```
+
+And `/tmp/segmentation-output.jpg` should contain the segmentation mask:
+
+![segmentation-output](g3doc/segmentation-output.png)
+
+[1]: https://tfhub.dev/google/lite-model/aiy/vision/classifier/birds_V1/3
+[2]: https://tfhub.dev/tensorflow/lite-model/ssd_mobilenet_v1/1/metadata/2
+[3]: https://tfhub.dev/tensorflow/lite-model/deeplabv3/1/metadata/2
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/g3doc/detection-output.png b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/detection-output.png
new file mode 100644
index 0000000..c8d56f4
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/detection-output.png
Binary files differ
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/g3doc/dogs.jpg b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/dogs.jpg
new file mode 100644
index 0000000..9db4bee
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/dogs.jpg
Binary files differ
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/g3doc/plane.jpg b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/plane.jpg
new file mode 100644
index 0000000..0edefa4
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/plane.jpg
Binary files differ
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/g3doc/segmentation-output.png b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/segmentation-output.png
new file mode 100644
index 0000000..e871df3
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/segmentation-output.png
Binary files differ
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/g3doc/sparrow.jpg b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/sparrow.jpg
new file mode 100644
index 0000000..25d213e
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/g3doc/sparrow.jpg
Binary files differ
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/image_classifier_demo.cc b/tensorflow_lite_support/examples/task/vision/desktop/image_classifier_demo.cc
new file mode 100644
index 0000000..cd97f4a
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/image_classifier_demo.cc
@@ -0,0 +1,173 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+// Example usage:
+// bazel run -c opt \
+//  tensorflow_lite_support/examples/task/vision/desktop:image_classifier_demo \
+//  -- \
+//  --model_path=/path/to/model.tflite \
+//  --image_path=/path/to/image.jpg
+
+#include <iostream>
+
+#include "absl/flags/flag.h"
+#include "absl/flags/parse.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/image_classifier.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/classifications_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/image_classifier_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+#include "tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.h"
+
+ABSL_FLAG(std::string, model_path, "",
+          "Absolute path to the '.tflite' image classifier model.");
+ABSL_FLAG(std::string, image_path, "",
+          "Absolute path to the image to classify. The image must be RGB or "
+          "RGBA (grayscale is not supported). The image EXIF orientation "
+          "flag, if any, is NOT taken into account.");
+ABSL_FLAG(int32, max_results, 5,
+          "Maximum number of classification results to display.");
+ABSL_FLAG(float, score_threshold, 0,
+          "Classification results with a confidence score below this value are "
+          "rejected. If >= 0, overrides the score threshold(s) provided in the "
+          "TFLite Model Metadata. Ignored otherwise.");
+ABSL_FLAG(
+    std::vector<std::string>, class_name_whitelist, {},
+    "Comma-separated list of class names that acts as a whitelist. If "
+    "non-empty, classification results whose 'class_name' is not in this list "
+    "are filtered out. Mutually exclusive with 'class_name_blacklist'.");
+ABSL_FLAG(
+    std::vector<std::string>, class_name_blacklist, {},
+    "Comma-separated list of class names that acts as a blacklist. If "
+    "non-empty, classification results whose 'class_name' is in this list "
+    "are filtered out. Mutually exclusive with 'class_name_whitelist'.");
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+ImageClassifierOptions BuildOptions() {
+  ImageClassifierOptions options;
+  options.mutable_model_file_with_metadata()->set_file_name(
+      absl::GetFlag(FLAGS_model_path));
+  options.set_max_results(absl::GetFlag(FLAGS_max_results));
+  if (absl::GetFlag(FLAGS_score_threshold) >= 0) {
+    options.set_score_threshold(absl::GetFlag(FLAGS_score_threshold));
+  }
+  for (const std::string& class_name :
+       absl::GetFlag(FLAGS_class_name_whitelist)) {
+    options.add_class_name_whitelist(class_name);
+  }
+  for (const std::string& class_name :
+       absl::GetFlag(FLAGS_class_name_blacklist)) {
+    options.add_class_name_blacklist(class_name);
+  }
+  return options;
+}
+
+void DisplayResult(const ClassificationResult& result) {
+  std::cout << "Results:\n";
+  for (int head = 0; head < result.classifications_size(); ++head) {
+    if (result.classifications_size() > 1) {
+      std::cout << absl::StrFormat(" Head index %d:\n", head);
+    }
+    const Classifications& classifications = result.classifications(head);
+    for (int rank = 0; rank < classifications.classes_size(); ++rank) {
+      const Class& classification = classifications.classes(rank);
+      std::cout << absl::StrFormat("  Rank #%d:\n", rank);
+      std::cout << absl::StrFormat("   index       : %d\n",
+                                   classification.index());
+      std::cout << absl::StrFormat("   score       : %.5f\n",
+                                   classification.score());
+      if (classification.has_class_name()) {
+        std::cout << absl::StrFormat("   class name  : %s\n",
+                                     classification.class_name());
+      }
+      if (classification.has_display_name()) {
+        std::cout << absl::StrFormat("   display name: %s\n",
+                                     classification.display_name());
+      }
+    }
+  }
+}
+
+absl::Status Classify() {
+  // Build ImageClassifier.
+  const ImageClassifierOptions& options = BuildOptions();
+  ASSIGN_OR_RETURN(std::unique_ptr<ImageClassifier> image_classifier,
+                   ImageClassifier::CreateFromOptions(options));
+
+  // Load image in a FrameBuffer.
+  ASSIGN_OR_RETURN(ImageData image,
+                   DecodeImageFromFile(absl::GetFlag(FLAGS_image_path)));
+  std::unique_ptr<FrameBuffer> frame_buffer;
+  if (image.channels == 3) {
+    frame_buffer =
+        CreateFromRgbRawBuffer(image.pixel_data, {image.width, image.height});
+  } else if (image.channels == 4) {
+    frame_buffer =
+        CreateFromRgbaRawBuffer(image.pixel_data, {image.width, image.height});
+  } else {
+    return absl::InvalidArgumentError(absl::StrFormat(
+        "Expected image with 3 (RGB) or 4 (RGBA) channels, found %d",
+        image.channels));
+  }
+
+  // Run classification and display results.
+  ASSIGN_OR_RETURN(ClassificationResult result,
+                   image_classifier->Classify(*frame_buffer));
+  DisplayResult(result);
+
+  // Cleanup and return.
+  ImageDataFree(&image);
+  return absl::OkStatus();
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+int main(int argc, char** argv) {
+  // Parse command line arguments and perform sanity checks.
+  absl::ParseCommandLine(argc, argv);
+  if (absl::GetFlag(FLAGS_model_path).empty()) {
+    std::cerr << "Missing mandatory 'model_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_image_path).empty()) {
+    std::cerr << "Missing mandatory 'image_path' argument.\n";
+    return 1;
+  }
+  if (!absl::GetFlag(FLAGS_class_name_whitelist).empty() &&
+      !absl::GetFlag(FLAGS_class_name_blacklist).empty()) {
+    std::cerr << "'class_name_whitelist' and 'class_name_blacklist' arguments "
+                 "are mutually exclusive.\n";
+    return 1;
+  }
+
+  // Run classification.
+  absl::Status status = tflite::task::vision::Classify();
+  if (status.ok()) {
+    return 0;
+  } else {
+    std::cerr << "Classification failed: " << status.message() << "\n";
+    return 1;
+  }
+}
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/image_segmenter_demo.cc b/tensorflow_lite_support/examples/task/vision/desktop/image_segmenter_demo.cc
new file mode 100644
index 0000000..02af824
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/image_segmenter_demo.cc
@@ -0,0 +1,201 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+// Example usage:
+// bazel run -c opt \
+//  tensorflow_lite_support/examples/task/vision/desktop:image_segmenter_demo \
+//  -- \
+//  --model_path=/path/to/model.tflite \
+//  --image_path=/path/to/image.jpg \
+//  --output_mask_png=/path/to/output/mask.png
+
+#include <iostream>
+
+#include "absl/flags/flag.h"
+#include "absl/flags/parse.h"
+#include "absl/status/status.h"
+#include "absl/strings/match.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/image_segmenter.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/segmentations_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+#include "tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.h"
+
+ABSL_FLAG(std::string, model_path, "",
+          "Absolute path to the '.tflite' image segmenter model.");
+ABSL_FLAG(std::string, image_path, "",
+          "Absolute path to the image to segment. The image must be RGB or "
+          "RGBA (grayscale is not supported). The image EXIF orientation "
+          "flag, if any, is NOT taken into account.");
+ABSL_FLAG(std::string, output_mask_png, "",
+          "Absolute path to the output category mask (confidence masks outputs "
+          "are not supported by this tool). Must have a '.png' extension.");
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+ImageSegmenterOptions BuildOptions() {
+  ImageSegmenterOptions options;
+  options.mutable_model_file_with_metadata()->set_file_name(
+      absl::GetFlag(FLAGS_model_path));
+  // Confidence masks are not supported by this tool: output_type is set to
+  // CATEGORY_MASK by default.
+  return options;
+}
+
+absl::Status EncodeMaskToPngFile(const SegmentationResult& result) {
+  if (result.segmentation_size() != 1) {
+    return absl::UnimplementedError(
+        "Image segmentation models with multiple output segmentations are not "
+        "supported by this tool.");
+  }
+  const Segmentation& segmentation = result.segmentation(0);
+  // Extract raw mask data as a uint8 pointer.
+  const uint8* raw_mask =
+      reinterpret_cast<const uint8*>(segmentation.category_mask().data());
+
+  // Create RgbImageData for the output mask.
+  uint8* pixel_data = static_cast<uint8*>(
+      malloc(segmentation.width() * segmentation.height() * 3 * sizeof(uint8)));
+  ImageData mask = {.pixel_data = pixel_data,
+                    .width = segmentation.width(),
+                    .height = segmentation.height(),
+                    .channels = 3};
+
+  // Populate RgbImageData from the raw mask and ColoredLabel-s.
+  for (int i = 0; i < segmentation.width() * segmentation.height(); ++i) {
+    Segmentation::ColoredLabel colored_label =
+        segmentation.colored_labels(raw_mask[i]);
+    pixel_data[3 * i] = colored_label.r();
+    pixel_data[3 * i + 1] = colored_label.g();
+    pixel_data[3 * i + 2] = colored_label.b();
+  }
+
+  // Encode mask as PNG.
+  RETURN_IF_ERROR(
+      EncodeImageToPngFile(mask, absl::GetFlag(FLAGS_output_mask_png)));
+  std::cout << absl::StrFormat("Category mask saved to: %s\n",
+                               absl::GetFlag(FLAGS_output_mask_png));
+
+  // Cleanup and return.
+  ImageDataFree(&mask);
+  return absl::OkStatus();
+}
+
+absl::Status DisplayColorLegend(const SegmentationResult& result) {
+  if (result.segmentation_size() != 1) {
+    return absl::UnimplementedError(
+        "Image segmentation models with multiple output segmentations are not "
+        "supported by this tool.");
+  }
+  const Segmentation& segmentation = result.segmentation(0);
+  const int num_labels = segmentation.colored_labels_size();
+
+  std::cout << "Color Legend:\n";
+  for (int index = 0; index < num_labels; ++index) {
+    Segmentation::ColoredLabel colored_label =
+        segmentation.colored_labels(index);
+    std::cout << absl::StrFormat(" (r: %03d, g: %03d, b: %03d):\n",
+                                 colored_label.r(), colored_label.g(),
+                                 colored_label.b());
+    std::cout << absl::StrFormat("  index       : %d\n", index);
+    if (colored_label.has_class_name()) {
+      std::cout << absl::StrFormat("  class name  : %s\n",
+                                   colored_label.class_name());
+    }
+    if (colored_label.has_display_name()) {
+      std::cout << absl::StrFormat("  display name: %s\n",
+                                   colored_label.display_name());
+    }
+  }
+  std::cout << "Tip: use a color picker on the output PNG file to inspect the "
+               "output mask with this legend.\n";
+
+  return absl::OkStatus();
+}
+
+absl::Status Segment() {
+  // Build ImageClassifier.
+  const ImageSegmenterOptions& options = BuildOptions();
+  ASSIGN_OR_RETURN(std::unique_ptr<ImageSegmenter> image_segmenter,
+                   ImageSegmenter::CreateFromOptions(options));
+
+  // Load image in a FrameBuffer.
+  ASSIGN_OR_RETURN(ImageData image,
+                   DecodeImageFromFile(absl::GetFlag(FLAGS_image_path)));
+  std::unique_ptr<FrameBuffer> frame_buffer;
+  if (image.channels == 3) {
+    frame_buffer =
+        CreateFromRgbRawBuffer(image.pixel_data, {image.width, image.height});
+  } else if (image.channels == 4) {
+    frame_buffer =
+        CreateFromRgbaRawBuffer(image.pixel_data, {image.width, image.height});
+  } else {
+    return absl::InvalidArgumentError(absl::StrFormat(
+        "Expected image with 3 (RGB) or 4 (RGBA) channels, found %d",
+        image.channels));
+  }
+
+  // Run segmentation and save category mask.
+  ASSIGN_OR_RETURN(SegmentationResult result,
+                   image_segmenter->Segment(*frame_buffer));
+  RETURN_IF_ERROR(EncodeMaskToPngFile(result));
+
+  // Display the legend.
+  RETURN_IF_ERROR(DisplayColorLegend(result));
+
+  // Cleanup and return.
+  ImageDataFree(&image);
+  return absl::OkStatus();
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+int main(int argc, char** argv) {
+  // Parse command line arguments and perform sanity checks.
+  absl::ParseCommandLine(argc, argv);
+  if (absl::GetFlag(FLAGS_model_path).empty()) {
+    std::cerr << "Missing mandatory 'model_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_image_path).empty()) {
+    std::cerr << "Missing mandatory 'image_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_output_mask_png).empty()) {
+    std::cerr << "Missing mandatory 'output_mask_png' argument.\n";
+    return 1;
+  }
+  if (!absl::EndsWithIgnoreCase(absl::GetFlag(FLAGS_output_mask_png), ".png")) {
+    std::cerr << "Argument 'output_mask_png' must end with '.png' or '.PNG'\n";
+    return 1;
+  }
+
+  // Run segmentation.
+  absl::Status status = tflite::task::vision::Segment();
+  if (status.ok()) {
+    return 0;
+  } else {
+    std::cerr << "Segmentation failed: " << status.message() << "\n";
+    return 1;
+  }
+}
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/object_detector_demo.cc b/tensorflow_lite_support/examples/task/vision/desktop/object_detector_demo.cc
new file mode 100644
index 0000000..b7ab651
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/object_detector_demo.cc
@@ -0,0 +1,251 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+// Example usage:
+// bazel run -c opt \
+//  tensorflow_lite_support/examples/task/vision/desktop:object_detector_demo \
+//  -- \
+//  --model_path=/path/to/model.tflite \
+//  --image_path=/path/to/image.jpg \
+//  --output_png=/path/to/output.png
+
+#include <iostream>
+#include <limits>
+
+#include "absl/flags/flag.h"
+#include "absl/flags/parse.h"
+#include "absl/status/status.h"
+#include "absl/strings/match.h"
+#include "absl/strings/str_format.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
+#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/object_detector.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/detections_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/object_detector_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+#include "tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.h"
+
+ABSL_FLAG(std::string, model_path, "",
+          "Absolute path to the '.tflite' object detector model.");
+ABSL_FLAG(std::string, image_path, "",
+          "Absolute path to the image to run detection on. The image must be "
+          "RGB or RGBA (grayscale is not supported). The image EXIF "
+          "orientation flag, if any, is NOT taken into account.");
+ABSL_FLAG(std::string, output_png, "",
+          "Absolute path to a file where to draw the detection results on top "
+          "of the input image. Must have a '.png' extension.");
+ABSL_FLAG(int32, max_results, 5,
+          "Maximum number of detection results to display.");
+ABSL_FLAG(
+    float, score_threshold, std::numeric_limits<float>::lowest(),
+    "Detection results with a confidence score below this value are "
+    "rejected. If specified, overrides the score threshold(s) provided in the "
+    "TFLite Model Metadata. Ignored otherwise.");
+ABSL_FLAG(
+    std::vector<std::string>, class_name_whitelist, {},
+    "Comma-separated list of class names that acts as a whitelist. If "
+    "non-empty, detections results whose 'class_name' is not in this list "
+    "are filtered out. Mutually exclusive with 'class_name_blacklist'.");
+ABSL_FLAG(std::vector<std::string>, class_name_blacklist, {},
+          "Comma-separated list of class names that acts as a blacklist. If "
+          "non-empty, detections results whose 'class_name' is in this list "
+          "are filtered out. Mutually exclusive with 'class_name_whitelist'.");
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+namespace {
+// The line thickness (in pixels) for drawing the detection results.
+constexpr int kLineThickness = 3;
+
+// The number of colors used for drawing the detection results.
+constexpr int kColorMapSize = 10;
+
+// The names of the colors used for drawing the detection results.
+constexpr std::array<absl::string_view, 10> kColorMapNames = {
+    "red",      "green",      "blue",      "yellow", "fuschia",
+    "dark red", "dark green", "dark blue", "gray",   "black"};
+
+// The colors used for drawing the detection results as a flattened array of
+// {R,G,B} components.
+constexpr uint8 kColorMapComponents[30] = {
+    255, 0, 0, 0, 255, 0, 0, 0, 255, 255, 255, 0,   255, 0, 255,
+    128, 0, 0, 0, 128, 0, 0, 0, 128, 128, 128, 128, 0,   0, 0};
+}  // namespace
+
+ObjectDetectorOptions BuildOptions() {
+  ObjectDetectorOptions options;
+  options.mutable_model_file_with_metadata()->set_file_name(
+      absl::GetFlag(FLAGS_model_path));
+  options.set_max_results(absl::GetFlag(FLAGS_max_results));
+  if (absl::GetFlag(FLAGS_score_threshold) >
+      std::numeric_limits<float>::lowest()) {
+    options.set_score_threshold(absl::GetFlag(FLAGS_score_threshold));
+  }
+  for (const std::string& class_name :
+       absl::GetFlag(FLAGS_class_name_whitelist)) {
+    options.add_class_name_whitelist(class_name);
+  }
+  for (const std::string& class_name :
+       absl::GetFlag(FLAGS_class_name_blacklist)) {
+    options.add_class_name_blacklist(class_name);
+  }
+  return options;
+}
+
+absl::Status EncodeResultToPngFile(const DetectionResult& result,
+                                   const ImageData* image) {
+  for (int index = 0; index < result.detections_size(); ++index) {
+    // Get bounding box as left, top, right, bottom.
+    const BoundingBox& box = result.detections(index).bounding_box();
+    const int left = box.origin_x();
+    const int top = box.origin_y();
+    const int right = box.origin_x() + box.width();
+    const int bottom = box.origin_y() + box.height();
+    // Get color components.
+    const uint8 r = kColorMapComponents[3 * (index % kColorMapSize)];
+    const uint8 g = kColorMapComponents[3 * (index % kColorMapSize) + 1];
+    const uint8 b = kColorMapComponents[3 * (index % kColorMapSize) + 2];
+    // Draw. Boxes might have coordinates outside of [0, w( x [0, h( so clamping
+    // is applied.
+    for (int y = std::max(0, top); y < std::min(image->height, bottom); ++y) {
+      for (int x = std::max(0, left); x < std::min(image->width, right); ++x) {
+        int pixel_index = image->channels * (image->width * y + x);
+        if (x < left + kLineThickness || x > right - kLineThickness ||
+            y < top + kLineThickness || y > bottom - kLineThickness) {
+          image->pixel_data[pixel_index] = r;
+          image->pixel_data[pixel_index + 1] = g;
+          image->pixel_data[pixel_index + 2] = b;
+        }
+      }
+    }
+  }
+  // Encode to PNG and return.
+  RETURN_IF_ERROR(
+      EncodeImageToPngFile(*image, absl::GetFlag(FLAGS_output_png)));
+  std::cout << absl::StrFormat("Results saved to: %s\n",
+                               absl::GetFlag(FLAGS_output_png));
+  return absl::OkStatus();
+}
+
+void DisplayResult(const DetectionResult& result) {
+  std::cout << "Results:\n";
+  for (int index = 0; index < result.detections_size(); ++index) {
+    std::cout << absl::StrFormat(" Detection #%d (%s):\n", index,
+                                 kColorMapNames[index % kColorMapSize]);
+    const Detection& detection = result.detections(index);
+    const BoundingBox& box = detection.bounding_box();
+    std::cout << absl::StrFormat("  Box: (x: %d, y: %d, w: %d, h: %d)\n",
+                                 box.origin_x(), box.origin_y(), box.width(),
+                                 box.height());
+    if (detection.classes_size() == 0) {
+      std::cout << "  No top-1 class available";
+    } else {
+      std::cout << "  Top-1 class:\n";
+      const Class& classification = detection.classes(0);
+      std::cout << absl::StrFormat("   index       : %d\n",
+                                   classification.index());
+      std::cout << absl::StrFormat("   score       : %.5f\n",
+                                   classification.score());
+      if (classification.has_class_name()) {
+        std::cout << absl::StrFormat("   class name  : %s\n",
+                                     classification.class_name());
+      }
+      if (classification.has_display_name()) {
+        std::cout << absl::StrFormat("   display name: %s\n",
+                                     classification.display_name());
+      }
+    }
+  }
+}
+
+absl::Status Detect() {
+  // Build ObjectDetector.
+  const ObjectDetectorOptions& options = BuildOptions();
+  ASSIGN_OR_RETURN(std::unique_ptr<ObjectDetector> object_detector,
+                   ObjectDetector::CreateFromOptions(options));
+
+  // Load image in a FrameBuffer.
+  ASSIGN_OR_RETURN(ImageData image,
+                   DecodeImageFromFile(absl::GetFlag(FLAGS_image_path)));
+  std::unique_ptr<FrameBuffer> frame_buffer;
+  if (image.channels == 3) {
+    frame_buffer =
+        CreateFromRgbRawBuffer(image.pixel_data, {image.width, image.height});
+  } else if (image.channels == 4) {
+    frame_buffer =
+        CreateFromRgbaRawBuffer(image.pixel_data, {image.width, image.height});
+  } else {
+    return absl::InvalidArgumentError(absl::StrFormat(
+        "Expected image with 3 (RGB) or 4 (RGBA) channels, found %d",
+        image.channels));
+  }
+
+  // Run object detection and draw results on input image.
+  ASSIGN_OR_RETURN(DetectionResult result,
+                   object_detector->Detect(*frame_buffer));
+  RETURN_IF_ERROR(EncodeResultToPngFile(result, &image));
+
+  // Display results as text.
+  DisplayResult(result);
+
+  // Cleanup and return.
+  ImageDataFree(&image);
+  return absl::OkStatus();
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+int main(int argc, char** argv) {
+  // Parse command line arguments and perform sanity checks.
+  absl::ParseCommandLine(argc, argv);
+  if (absl::GetFlag(FLAGS_model_path).empty()) {
+    std::cerr << "Missing mandatory 'model_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_image_path).empty()) {
+    std::cerr << "Missing mandatory 'image_path' argument.\n";
+    return 1;
+  }
+  if (absl::GetFlag(FLAGS_output_png).empty()) {
+    std::cerr << "Missing mandatory 'output_png' argument.\n";
+    return 1;
+  }
+  if (!absl::EndsWithIgnoreCase(absl::GetFlag(FLAGS_output_png), ".png")) {
+    std::cerr << "Argument 'output_png' must end with '.png' or '.PNG'\n";
+    return 1;
+  }
+  if (!absl::GetFlag(FLAGS_class_name_whitelist).empty() &&
+      !absl::GetFlag(FLAGS_class_name_blacklist).empty()) {
+    std::cerr << "'class_name_whitelist' and 'class_name_blacklist' arguments "
+                 "are mutually exclusive.\n";
+    return 1;
+  }
+
+  // Run detection.
+  absl::Status status = tflite::task::vision::Detect();
+  if (status.ok()) {
+    return 0;
+  } else {
+    std::cerr << "Detection failed: " << status.message() << "\n";
+    return 1;
+  }
+}
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/utils/BUILD b/tensorflow_lite_support/examples/task/vision/desktop/utils/BUILD
new file mode 100644
index 0000000..9a837e2
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/utils/BUILD
@@ -0,0 +1,22 @@
+package(
+    default_visibility = [
+        "//tensorflow_lite_support:users",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "image_utils",
+    srcs = ["image_utils.cc"],
+    hdrs = ["image_utils.h"],
+    deps = [
+        "//tensorflow_lite_support/cc/port:integral_types",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@stblib//:stb_image",
+        "@stblib//:stb_image_write",
+    ],
+)
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.cc b/tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.cc
new file mode 100644
index 0000000..7c4604e
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.cc
@@ -0,0 +1,94 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.h"
+
+#include <cstdlib>
+#include <cstring>
+#include <vector>
+
+// These need to be defined for stb_image.h and stb_image_write.h to include
+// the actual implementations of image decoding/encoding functions.
+#define STB_IMAGE_IMPLEMENTATION
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+
+#include "absl/status/status.h"
+#include "absl/strings/match.h"
+#include "absl/strings/str_format.h"
+#include "stb_image.h"
+#include "stb_image_write.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+using ::tflite::support::StatusOr;
+
+StatusOr<ImageData> DecodeImageFromFile(const std::string& file_name) {
+  ImageData image_data;
+  image_data.pixel_data = stbi_load(file_name.c_str(), &image_data.width,
+                                    &image_data.height, &image_data.channels,
+                                    /*desired_channels=*/0);
+  if (image_data.pixel_data == nullptr) {
+    return absl::InternalError(absl::StrFormat(
+        "An error occurred while decoding image: %s", stbi_failure_reason()));
+  }
+  if (image_data.channels != 1 && image_data.channels != 3 &&
+      image_data.channels != 4) {
+    stbi_image_free(image_data.pixel_data);
+    return absl::UnimplementedError(
+        absl::StrFormat("Expected image with 1 (grayscale), 3 (RGB) or 4 "
+                        "(RGBA) channels, found %d",
+                        image_data.channels));
+  }
+  return image_data;
+}
+
+absl::Status EncodeImageToPngFile(const ImageData& image_data,
+                                  const std::string& image_path) {
+  // Sanity check inputs.
+  if (image_data.width <= 0 || image_data.height <= 0) {
+    return absl::InvalidArgumentError(
+        absl::StrFormat("Expected positive image dimensions, found %d x %d.",
+                        image_data.width, image_data.height));
+  }
+  if (image_data.channels != 1 && image_data.channels != 3 &&
+      image_data.channels != 4) {
+    return absl::UnimplementedError(
+        absl::StrFormat("Expected image data with 1 (grayscale), 3 (RGB) or 4 "
+                        "(RGBA) channels, found %d",
+                        image_data.channels));
+  }
+  if (image_data.pixel_data == nullptr) {
+    return absl::InvalidArgumentError(
+        "Expected pixel data to be set, found nullptr.");
+  }
+
+  if (stbi_write_png(
+          image_path.c_str(), image_data.width, image_data.height,
+          image_data.channels, image_data.pixel_data,
+          /*stride_in_bytes=*/image_data.width * image_data.channels) == 0) {
+    return absl::InternalError("An error occurred while encoding image.");
+  }
+
+  return absl::OkStatus();
+}
+
+void ImageDataFree(ImageData* image) { stbi_image_free(image->pixel_data); }
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.h b/tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.h
new file mode 100644
index 0000000..38f62b6
--- /dev/null
+++ b/tensorflow_lite_support/examples/task/vision/desktop/utils/image_utils.h
@@ -0,0 +1,58 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_EXAMPLES_TASK_VISION_DESKTOP_UTILS_IMAGE_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_EXAMPLES_TASK_VISION_DESKTOP_UTILS_IMAGE_UTILS_H_
+
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow_lite_support/cc/port/integral_types.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Image data with pixels stored as a row-major flattened array.
+// Channels can be:
+// 1 : grayscale
+// 3 : RGB, interleaved
+// 4 : RGBA, interleaved
+struct ImageData {
+  uint8* pixel_data;
+  int width;
+  int height;
+  int channels;
+};
+
+// Decodes image file and returns the corresponding image if no error
+// occurred. If decoding succeeded, the caller must manage deletion of the
+// underlying pixel data using `ImageDataFree`.
+// Supports a wide range of image formats, listed in `stb_image/stb_image.h`.
+tflite::support::StatusOr<ImageData> DecodeImageFromFile(
+    const std::string& file_name);
+
+// Encodes the image provided as an ImageData as lossless PNG to the provided
+// path.
+absl::Status EncodeImageToPngFile(const ImageData& image_data,
+                                  const std::string& image_path);
+
+// Releases image pixel data memory.
+void ImageDataFree(ImageData* image);
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_EXAMPLES_TASK_VISION_DESKTOP_UTILS_IMAGE_UTILS_H_
diff --git a/tensorflow_lite_support/ios/BUILD b/tensorflow_lite_support/ios/BUILD
new file mode 100644
index 0000000..07b8951
--- /dev/null
+++ b/tensorflow_lite_support/ios/BUILD
@@ -0,0 +1,48 @@
+# TensorFlow Lite Task Library - Text
+
+load(
+    "@org_tensorflow//tensorflow/lite/experimental/ios:ios.bzl",
+    "TFL_MINIMUM_OS_VERSION",
+    "tflite_ios_static_framework",
+)
+load(
+    "//tensorflow_lite_support/ios:ios.bzl",
+    "strip_c_api_include_path_prefix",
+)
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+strip_c_api_include_path_prefix(
+    name = "strip_c_api_include_path",
+    hdr_labels = [
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:bert_nl_classifier_c_api.h",
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier_c_api.h",
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier_c_api_common.h",
+        "//tensorflow_lite_support/cc/task/text/qa:bert_qa_c_api.h",
+    ],
+)
+
+# This target builds a monolithic static framework for the TFLite Text API,
+# which includes the TFLite runtime in it.
+#
+# bazel build -c opt --config=ios_fat //tensorflow_lite_support/ios:TensorFlowLiteTaskTextC_framework
+tflite_ios_static_framework(
+    name = "TensorFlowLiteTaskTextC_framework",
+    hdrs = [
+        ":bert_nl_classifier_c_api.h",
+        ":bert_qa_c_api.h",
+        ":nl_classifier_c_api.h",
+        ":nl_classifier_c_api_common.h",
+    ],
+    allowlist_symbols_file = ":allowlist_TensorFlowLiteTaskText.txt",
+    bundle_name = "TensorFlowLiteTaskTextC",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:bert_nl_classifier_c_api",
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier_c_api",
+        "//tensorflow_lite_support/cc/task/text/qa:bert_qa_c_api",
+    ],
+)
diff --git a/tensorflow_lite_support/ios/TensorFlowLiteTaskText.podspec.template b/tensorflow_lite_support/ios/TensorFlowLiteTaskText.podspec.template
new file mode 100644
index 0000000..62c3f33
--- /dev/null
+++ b/tensorflow_lite_support/ios/TensorFlowLiteTaskText.podspec.template
@@ -0,0 +1,44 @@
+Pod::Spec.new do |s|
+  s.name             = 'TensorFlowLiteTaskText'
+  s.version          = '${TFLS_BUILD_VERSION}'
+  s.authors          = 'Google Inc.'
+  s.license          = { :type => 'Apache' }
+  s.homepage         = 'https://github.com/tensorflow/tflite-support'
+  s.source           = { :http => '${TFLS_DOWNLOAD_URL}' }
+  s.summary          = 'TensorFlow Lite Task Library - Text'
+  s.description      = 'The Natural Language APIs of the TFLite Task Library'
+
+  s.ios.deployment_target = '9.0'
+
+  s.module_name = 'TensorFlowLiteTaskText'
+  s.static_framework = true
+
+  s.dependency 'GoogleToolboxForMac', '2.2.1'
+
+  objc_dir = 'tensorflow_lite_support/ios/task/text/'
+  s.public_header_files = [
+    objc_dir + 'apis/*.h',
+    objc_dir + '{nlclassifier,qa}/Sources/*.h'
+  ]
+
+  cc_dir = 'tensorflow_lite_support/cc/task/text/'
+  s.source_files = [
+    cc_dir + '{nlclassifier,qa}/*_c_api*.h',
+    objc_dir + 'apis/*.h',
+    objc_dir + '{nlclassifier,qa}/Sources/*.{h,m,mm}'
+  ]
+  s.module_map = objc_dir + 'apis/framework.modulemap'
+  s.pod_target_xcconfig = {
+    'HEADER_SEARCH_PATHS' =>
+      '"${PODS_TARGET_SRCROOT}" ' +
+      '"${PODS_TARGET_SRCROOT}/' + cc_dir + 'nlclassifier" ' +
+      '"${PODS_TARGET_SRCROOT}/' + cc_dir + 'qa" ' +
+      '"${PODS_TARGET_SRCROOT}/' + objc_dir + 'apis" ' +
+      '"${PODS_TARGET_SRCROOT}/' + objc_dir + 'nlclassifier/Sources" ' +
+      '"${PODS_TARGET_SRCROOT}/' + objc_dir + 'qa/Sources"',
+    'VALID_ARCHS' => 'x86_64 armv7 arm64',
+  }
+
+  s.library = 'c++'
+  s.vendored_frameworks = 'Frameworks/TensorFlowLiteTaskTextC.framework'
+end
diff --git a/tensorflow_lite_support/ios/allowlist_TensorFlowLiteTaskText.txt b/tensorflow_lite_support/ios/allowlist_TensorFlowLiteTaskText.txt
new file mode 100644
index 0000000..3af5b0b
--- /dev/null
+++ b/tensorflow_lite_support/ios/allowlist_TensorFlowLiteTaskText.txt
@@ -0,0 +1,3 @@
+_NLClassifier*
+_BertNLClassifier*
+_BertQuestionAnswerer*
diff --git a/tensorflow_lite_support/ios/ios.bzl b/tensorflow_lite_support/ios/ios.bzl
new file mode 100644
index 0000000..cb8c92a
--- /dev/null
+++ b/tensorflow_lite_support/ios/ios.bzl
@@ -0,0 +1,30 @@
+"""TensorFlow Lite Support Library Helper Rules for iOS"""
+
+# When the static framework is built with bazel, the all header files are moved
+# to the "Headers" directory with no header path prefixes. This auxiliary rule
+# is used for stripping the path prefix to the C API header files included by
+# other C API header files.
+def strip_c_api_include_path_prefix(name, hdr_labels, prefix = ""):
+    """Create modified header files with the common.h include path stripped out.
+
+    Args:
+      name: The name to be used as a prefix to the generated genrules.
+      hdr_labels: List of header labels to strip out the include path. Each
+          label must end with a colon followed by the header file name.
+      prefix: Optional prefix path to prepend to the header inclusion path.
+    """
+
+    for hdr_label in hdr_labels:
+        hdr_filename = hdr_label.split(":")[-1]
+        hdr_basename = hdr_filename.split(".")[0]
+
+        native.genrule(
+            name = "{}_{}".format(name, hdr_basename),
+            srcs = [hdr_label],
+            outs = [hdr_filename],
+            cmd = """
+            sed 's|#include ".*/\\([^/]\\{{1,\\}}\\.h\\)"|#include "{}\\1"|'\
+            "$(location {})"\
+            > "$@"
+            """.format(prefix, hdr_label),
+        )
diff --git a/tensorflow_lite_support/ios/task/text/apis/TFLTaskText.h b/tensorflow_lite_support/ios/task/text/apis/TFLTaskText.h
new file mode 100644
index 0000000..a42a4b3
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/apis/TFLTaskText.h
@@ -0,0 +1,17 @@
+// Copyright 2020 Google Inc. All rights reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at:
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import "TFLBertNLClassifier.h"
+#import "TFLBertQuestionAnswerer.h"
+#import "TFLNLClassifier.h"
diff --git a/tensorflow_lite_support/ios/task/text/apis/framework.modulemap b/tensorflow_lite_support/ios/task/text/apis/framework.modulemap
new file mode 100644
index 0000000..3e26762
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/apis/framework.modulemap
@@ -0,0 +1,4 @@
+framework module TensorFlowLiteTaskText {
+    umbrella header "TFLTaskText.h"
+    export *
+}
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/BUILD b/tensorflow_lite_support/ios/task/text/nlclassifier/BUILD
new file mode 100644
index 0000000..fb369e9
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/BUILD
@@ -0,0 +1,125 @@
+load("@build_bazel_rules_swift//swift:swift.bzl", "swift_library")
+load("@org_tensorflow//tensorflow/lite/experimental/ios:ios.bzl", "TFL_DEFAULT_TAGS", "TFL_DISABLED_SANITIZER_TAGS", "TFL_MINIMUM_OS_VERSION")
+load("@build_bazel_rules_apple//apple:ios.bzl", "ios_unit_test")
+load("@org_tensorflow//tensorflow/lite:special_rules.bzl", "tflite_ios_lab_runner")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+objc_library(
+    name = "TFLBertNLClassifier",
+    srcs = ["Sources/TFLBertNLClassifier.m"],
+    hdrs = ["Sources/TFLBertNLClassifier.h"],
+    module_name = "TFLBertNLClassifier",
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:bert_nl_classifier_c_api",
+        "@google_toolbox_for_mac//:GTM_Defines",
+    ],
+)
+
+swift_library(
+    name = "TFLBertNLClassifierSwiftTestLibrary",
+    testonly = 1,
+    srcs = ["Tests/TFLBertNLClassifierTest.swift"],
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:nl_classifier_models",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLBertNLClassifier",
+        "//third_party/swift/xctest",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLBertNLClassifierSwiftTest",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLBertNLClassifierSwiftTestLibrary",
+    ],
+)
+
+objc_library(
+    name = "TFLBertNLClassifierObjcTestLibrary",
+    testonly = 1,
+    srcs = ["Tests/TFLBertNLClassifierTest.m"],
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:nl_classifier_models",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLBertNLClassifier",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLBertNLClassifierObjcTest",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLBertNLClassifierObjcTestLibrary",
+    ],
+)
+
+objc_library(
+    name = "TFLNLClassifier",
+    srcs = ["Sources/TFLNLClassifier.m"],
+    hdrs = ["Sources/TFLNLClassifier.h"],
+    module_name = "TFLNLClassifier",
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier_c_api",
+        "@google_toolbox_for_mac//:GTM_Defines",
+    ],
+)
+
+swift_library(
+    name = "TFLNLClassifierSwiftTestLibrary",
+    testonly = 1,
+    srcs = ["Tests/TFLNLClassifierTest.swift"],
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:nl_classifier_models",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLNLClassifier",
+        "//third_party/swift/xctest",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLNLClassifierSwiftTest",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLNLClassifierSwiftTestLibrary",
+    ],
+)
+
+objc_library(
+    name = "TFLNLClassifierObjcTestLibrary",
+    testonly = 1,
+    srcs = ["Tests/TFLNLClassifierTest.m"],
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:nl_classifier_models",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLNLClassifier",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLNLClassifierObjcTest",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLNLClassifierObjcTestLibrary",
+    ],
+)
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.h b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.h
new file mode 100644
index 0000000..ceed6fa
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.h
@@ -0,0 +1,51 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import <Foundation/Foundation.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+/**
+ * Classifier API for NLClassification tasks with Bert models, categorizes string into different
+ * classes. The API expects a Bert based TFLite model with metadata populated.
+ *
+ * The metadata should contain the following information:
+ *   1 input_process_unit for Wordpiece/Sentencepiece Tokenizer.
+ *   3 input tensors with names "ids", "mask" and "segment_ids".
+ *   1 output tensor of type float32[1, 2], with a optionally attached label file. If a label
+ *     file is attached, the file should be a plain text file with one label per line, the number
+ *     of labels should match the number of categories the model outputs.
+ */
+@interface TFLBertNLClassifier : NSObject
+
+/**
+ * Creates TFLBertNLClassifier from a model file.
+ *
+ * @param modelPath Path to the classification model.
+ * @return A TFLBertNLClassifier instance.
+ */
++ (instancetype)bertNLClassifierWithModelPath:(NSString *)modelPath
+    NS_SWIFT_NAME(bertNLClassifier(modelPath:));
+
+/**
+ * Performs classification on a NSString input, returns <NSString *, NSNumber *>
+ * for categories and socres.
+ *
+ * @param text input text to the model.
+ * @return A NSDictionary of categorization results.
+ */
+- (NSDictionary<NSString *, NSNumber *> *)classifyWithText:(NSString *)text
+    NS_SWIFT_NAME(classify(text:));
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.m b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.m
new file mode 100644
index 0000000..24c37b7
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.m
@@ -0,0 +1,60 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.h"
+#import "GTMDefines.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier_c_api.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h"
+
+NS_ASSUME_NONNULL_BEGIN
+
+@interface TFLBertNLClassifier ()
+/** BertNLClassifier backed by C API */
+@property(nonatomic) BertNLClassifier *bertNLClassifier;
+@end
+
+@implementation TFLBertNLClassifier
+
+- (void)dealloc {
+  BertNLClassifierDelete(_bertNLClassifier);
+}
+
++ (instancetype)bertNLClassifierWithModelPath:(NSString *)modelPath {
+  BertNLClassifier *classifier = BertNLClassifierFromFile(modelPath.UTF8String);
+
+  _GTMDevAssert(classifier, @"Failed to create BertNLClassifier");
+  return [[TFLBertNLClassifier alloc] initWithBertNLClassifier:classifier];
+}
+
+- (instancetype)initWithBertNLClassifier:(BertNLClassifier *)bertNLClassifier {
+  self = [super init];
+  if (self) {
+    _bertNLClassifier = bertNLClassifier;
+  }
+  return self;
+}
+
+- (NSDictionary<NSString *, NSNumber *> *)classifyWithText:(NSString *)text {
+  struct Categories *cCategories = BertNLClassifierClassify(_bertNLClassifier, text.UTF8String);
+  NSMutableDictionary<NSString *, NSNumber *> *ret = [NSMutableDictionary dictionary];
+  for (int i = 0; i < cCategories->size; i++) {
+    struct Category cCategory = cCategories->categories[i];
+    [ret setValue:[NSNumber numberWithDouble:cCategory.score]
+           forKey:[NSString stringWithUTF8String:cCategory.text]];
+  }
+  NLClassifierCategoriesDelete(cCategories);
+  return ret;
+}
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.h b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.h
new file mode 100644
index 0000000..ceb8d2e
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.h
@@ -0,0 +1,86 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import <Foundation/Foundation.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+/**
+ * Options to identify input and output tensors of the model.
+ */
+@interface TFLNLClassifierOptions : NSObject
+@property(nonatomic) int inputTensorIndex;
+@property(nonatomic) int outputScoreTensorIndex;
+@property(nonatomic) int outputLabelTensorIndex;
+@property(nonatomic) NSString *inputTensorName;
+@property(nonatomic) NSString *outputScoreTensorName;
+@property(nonatomic) NSString *outputLabelTensorName;
+@end
+
+/**
+ * Classifier API for natural language classification tasks, categorizes string into different
+ * classes.
+ *
+ * The API expects a TFLite model with the following input/output tensor:
+ *
+ *   Input tensor (kTfLiteString)
+ *     input of the model, accepts a string.
+ *
+ *   Output score tensor
+ *     (kTfLiteUInt8/kTfLiteInt8/kTfLiteInt16/kTfLiteFloat32/kTfLiteFloat64/kTfLiteBool)
+ *     output scores for each class, if type is one of the Int types, dequantize it, if it
+ *       is Bool type, convert the values to 0.0 and 1.0 respectively.
+ *
+ *     can have an optional associated file in metadata for labels, the file should be a
+ *       plain text file with one label per line, the number of labels should match the number
+ *       of categories the model outputs. Output label tensor: optional (kTfLiteString) -
+ *       output classname for each class, should be of the same length with scores. If this
+ *       tensor is not present, the API uses score indices as classnames. - will be ignored if
+ *       output score tensor already has an associated label file.
+ *
+ *   Optional Output label tensor (kTfLiteString/kTfLiteInt32)
+ *     output classname for each class, should be of the same length with scores. If this
+ *       tensor is not present, the API uses score indices as classnames.
+ *
+ *     will be ignored if output score tensor already has an associated labe file.
+ *
+ * By default the API tries to find the input/output tensors with default configurations in
+ * TFLNLClassifierOptions, with tensor name prioritized over tensor index. The option is
+ * configurable for different TFLite models.
+ */
+@interface TFLNLClassifier : NSObject
+
+/**
+ * Creates a TFLNLClassifier instance from TFLNLClassifierOptions.
+ *
+ * @param modelPath The file path to the tflite mdoel.
+ * @param options The TFLNLClassifierOptions to configure the model.
+ *
+ * @return A TFLNLClassifier instance.
+ */
++ (instancetype)nlClassifierWithModelPath:(NSString *)modelPath
+                                  options:(TFLNLClassifierOptions *)options
+    NS_SWIFT_NAME(nlClassifier(modelPath:options:));
+
+/**
+ * Performs classification on a NSString input, returns <NSString *, NSNumber *>
+ * for categories and socres.
+ *
+ * @param text input text to the model.
+ * @return A NSDictionary of categorization results.
+ */
+- (NSDictionary<NSString *, NSNumber *> *)classifyWithText:(NSString *)text
+    NS_SWIFT_NAME(classify(text:));
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.m b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.m
new file mode 100644
index 0000000..01a4818
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.m
@@ -0,0 +1,79 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.h"
+#import "GTMDefines.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier_c_api_common.h"
+
+NS_ASSUME_NONNULL_BEGIN
+
+@implementation TFLNLClassifierOptions
+@synthesize inputTensorIndex;
+@synthesize outputScoreTensorIndex;
+@synthesize outputLabelTensorIndex;
+@synthesize inputTensorName;
+@synthesize outputScoreTensorName;
+@synthesize outputLabelTensorName;
+@end
+
+@interface TFLNLClassifier ()
+/** NLClassifier backed by C API */
+@property(nonatomic) NLClassifier *nlClassifier;
+@end
+
+@implementation TFLNLClassifier
+
+- (void)dealloc {
+  NLClassifierDelete(_nlClassifier);
+}
+
++ (instancetype)nlClassifierWithModelPath:(NSString *)modelPath
+                                  options:(TFLNLClassifierOptions *)options {
+  struct NLClassifierOptions cOptions = {
+          .input_tensor_index = options.inputTensorIndex,
+          .output_score_tensor_index = options.outputScoreTensorIndex,
+          .output_label_tensor_index = options.outputLabelTensorIndex,
+          .input_tensor_name = options.inputTensorName.UTF8String,
+          .output_score_tensor_name =
+              options.outputScoreTensorName.UTF8String,
+          .output_label_tensor_name =
+              options.outputLabelTensorName.UTF8String
+  };
+  NLClassifier *classifier = NLClassifierFromFileAndOptions(modelPath.UTF8String, &cOptions);
+  _GTMDevAssert(classifier, @"Failed to create NLClassifier");
+  return [[TFLNLClassifier alloc] initWithNLClassifier:classifier];
+}
+
+- (instancetype)initWithNLClassifier:(NLClassifier *)nlClassifier {
+  self = [super init];
+  if (self) {
+    _nlClassifier = nlClassifier;
+  }
+  return self;
+}
+
+- (NSDictionary<NSString *, NSNumber *> *)classifyWithText:(NSString *)text {
+  struct Categories *cCategories = NLClassifierClassify(_nlClassifier, text.UTF8String);
+  NSMutableDictionary<NSString *, NSNumber *> *ret = [NSMutableDictionary dictionary];
+  for (int i = 0; i < cCategories->size; i++) {
+    struct Category cCategory = cCategories->categories[i];
+    [ret setValue:[NSNumber numberWithDouble:cCategory.score]
+           forKey:[NSString stringWithUTF8String:cCategory.text]];
+  }
+  NLClassifierCategoriesDelete(cCategories);
+  return ret;
+}
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLBertNLClassifierTest.m b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLBertNLClassifierTest.m
new file mode 100644
index 0000000..9565bfb
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLBertNLClassifierTest.m
@@ -0,0 +1,61 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLBertNLClassifier.h"
+
+#import <XCTest/XCTest.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+@interface TFLBertNLClassifierTest : XCTestCase
+@property(nonatomic, nullable) NSString *bertModelPath;
+@end
+
+@implementation TFLBertNLClassifierTest
+#pragma mark - Tests
+
+- (void)setUp {
+  [super setUp];
+  NSBundle *bundle = [NSBundle bundleForClass:[self class]];
+  self.bertModelPath = [bundle pathForResource:@"test_model_nl_classifier_bert"
+                                        ofType:@"tflite"];
+}
+
+- (void)testClassifyPositiveResult {
+  TFLBertNLClassifier* bertNLClassifier =
+    [TFLBertNLClassifier bertNLClassifierWithModelPath:self.bertModelPath];
+
+  XCTAssertNotNil(bertNLClassifier);
+
+  NSDictionary<NSString *, NSNumber *> * categories =
+    [bertNLClassifier classifyWithText:@"it's a charming and often affecting journey"];
+
+  XCTAssertGreaterThan([categories[@"positive"] doubleValue],
+                       [categories[@"negative"] doubleValue]);
+}
+
+- (void)testClassifyNegativeResult {
+  TFLBertNLClassifier* bertNLClassifier =
+    [TFLBertNLClassifier bertNLClassifierWithModelPath:self.bertModelPath];
+
+  XCTAssertNotNil(bertNLClassifier);
+
+  NSDictionary<NSString *, NSNumber *> * categories =
+    [bertNLClassifier classifyWithText:@"unflinchingly bleak and desperate"];
+
+  XCTAssertGreaterThan([categories[@"negative"] doubleValue],
+                       [categories[@"positive"] doubleValue]);
+}
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLBertNLClassifierTest.swift b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLBertNLClassifierTest.swift
new file mode 100644
index 0000000..d331b04
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLBertNLClassifierTest.swift
@@ -0,0 +1,45 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+import XCTest
+
+@testable import TFLBertNLClassifier
+
+class TFLBertNLClassifierTest: XCTestCase {
+
+  static let bundle = Bundle(for: TFLBertNLClassifierTest.self)
+  static let bertModelPath = bundle.path(forResource: "test_model_nl_classifier_bert", ofType: "tflite")!
+
+  func testClassifyPositiveResult() {
+    let bertNLClassifier = TFLBertNLClassifier.bertNLClassifier(
+      modelPath: TFLBertNLClassifierTest.bertModelPath)
+
+    XCTAssertNotNil(bertNLClassifier)
+
+    let categories = bertNLClassifier.classify(text: "it's a charming and often affecting journey")
+
+    XCTAssertGreaterThan(categories["positive"]!.doubleValue, categories["negative"]!.doubleValue)
+  }
+
+  func testClassifyNegativeResult() {
+    let bertNLClassifier = TFLBertNLClassifier.bertNLClassifier(
+      modelPath: TFLBertNLClassifierTest.bertModelPath)
+
+    XCTAssertNotNil(bertNLClassifier)
+
+    let categories = bertNLClassifier.classify(text: "unflinchingly bleak and desperate")
+
+    XCTAssertGreaterThan(categories["negative"]!.doubleValue, categories["positive"]!.doubleValue)
+  }
+}
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLNLClassifierTest.m b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLNLClassifierTest.m
new file mode 100644
index 0000000..40814ac
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLNLClassifierTest.m
@@ -0,0 +1,65 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "tensorflow_lite_support/ios/task/text/nlclassifier/Sources/TFLNLClassifier.h"
+
+#import <XCTest/XCTest.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+@interface TFLNLClassifierTest : XCTestCase
+@property(nonatomic, nullable) NSString *modelPath;
+@property(nonatomic, nullable) TFLNLClassifierOptions *modelOptions;
+@end
+
+@implementation TFLNLClassifierTest
+#pragma mark - Tests
+
+- (void)setUp {
+  [super setUp];
+  NSBundle *bundle = [NSBundle bundleForClass:[self class]];
+  self.modelPath = [bundle pathForResource:@"test_model_nl_classifier_with_regex_tokenizer"
+                                    ofType:@"tflite"];
+  self.modelOptions = [[TFLNLClassifierOptions alloc] init];
+  [self.modelOptions setInputTensorName:@"input_text"];
+  [self.modelOptions setOutputScoreTensorName:@"probability"];
+}
+
+- (void)testClassifyPositiveResult {
+  TFLNLClassifier *nlClassifier = [TFLNLClassifier nlClassifierWithModelPath:self.modelPath
+                                                                     options:self.modelOptions];
+
+  XCTAssertNotNil(nlClassifier);
+
+  NSDictionary<NSString *, NSNumber *> *categories = [nlClassifier
+      classifyWithText:@"This is the best movie I’ve seen in recent years. Strongly recommend it!"];
+
+  XCTAssertGreaterThan([categories[@"Positive"] doubleValue],
+                       [categories[@"Negative"] doubleValue]);
+}
+
+- (void)testClassifyNegativeResult {
+  TFLNLClassifier *nlClassifier = [TFLNLClassifier nlClassifierWithModelPath:self.modelPath
+                                                                     options:self.modelOptions];
+
+  XCTAssertNotNil(nlClassifier);
+
+  NSDictionary<NSString *, NSNumber *> *categories =
+      [nlClassifier classifyWithText:@"What a waste of my time."];
+
+  XCTAssertGreaterThan([categories[@"Negative"] doubleValue],
+                       [categories[@"Positive"] doubleValue]);
+}
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLNLClassifierTest.swift b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLNLClassifierTest.swift
new file mode 100644
index 0000000..fb80e5d
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/nlclassifier/Tests/TFLNLClassifierTest.swift
@@ -0,0 +1,58 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+import XCTest
+
+@testable import TFLNLClassifier
+
+class TFLNLClassifierTest: XCTestCase {
+
+  static let bundle = Bundle(for: TFLNLClassifierTest.self)
+  static let modelPath = bundle.path(
+    forResource: "test_model_nl_classifier_with_regex_tokenizer",
+    ofType: "tflite")!
+
+  var modelOptions:TFLNLClassifierOptions!;
+
+  override func setUp() {
+    modelOptions = TFLNLClassifierOptions()
+    modelOptions.inputTensorName = "input_text"
+    modelOptions.outputScoreTensorName = "probability"
+  }
+
+  func testClassifyPositiveResult() {
+    let nlClassifier = TFLNLClassifier.nlClassifier(
+      modelPath: TFLNLClassifierTest.modelPath,
+      options: modelOptions)
+
+    XCTAssertNotNil(nlClassifier)
+
+    let categories = nlClassifier.classify(
+      text: "This is the best movie I’ve seen in recent years. Strongly recommend it!")
+
+    XCTAssertGreaterThan(categories["Positive"]!.doubleValue, categories["Negative"]!.doubleValue)
+  }
+
+   func testClassifyNegativeResult() {
+     let nlClassifier = TFLNLClassifier.nlClassifier(
+       modelPath: TFLNLClassifierTest.modelPath,
+       options: modelOptions)
+
+     XCTAssertNotNil(nlClassifier)
+
+     let categories = nlClassifier.classify(text: "What a waste of my time.")
+
+     XCTAssertGreaterThan(categories["Negative"]!.doubleValue, categories["Positive"]!.doubleValue)
+   }
+}
diff --git a/tensorflow_lite_support/ios/task/text/qa/BUILD b/tensorflow_lite_support/ios/task/text/qa/BUILD
new file mode 100644
index 0000000..7998a8e
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/qa/BUILD
@@ -0,0 +1,71 @@
+load("@build_bazel_rules_swift//swift:swift.bzl", "swift_library")
+load("@org_tensorflow//tensorflow/lite/experimental/ios:ios.bzl", "TFL_DEFAULT_TAGS", "TFL_DISABLED_SANITIZER_TAGS", "TFL_MINIMUM_OS_VERSION")
+load("@build_bazel_rules_apple//apple:ios.bzl", "ios_unit_test")
+load("@org_tensorflow//tensorflow/lite:special_rules.bzl", "tflite_ios_lab_runner")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+objc_library(
+    name = "TFLBertQuestionAnswerer",
+    srcs = ["Sources/TFLBertQuestionAnswerer.m"],
+    hdrs = ["Sources/TFLBertQuestionAnswerer.h"],
+    module_name = "TFLBertQuestionAnswerer",
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/qa:bert_qa_c_api",
+        "@google_toolbox_for_mac//:GTM_Defines",
+    ],
+)
+
+swift_library(
+    name = "TFLBertQuestionAnswererSwiftTestLibrary",
+    testonly = 1,
+    srcs = glob(["Tests/*.swift"]),
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:albert_model",
+        "//tensorflow_lite_support/cc/test/testdata/task/text:mobile_bert_model",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLBertQuestionAnswerer",
+        "//third_party/swift/xctest",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLBertQuestionAnswererSwiftTest",
+    size = "large",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLBertQuestionAnswererSwiftTestLibrary",
+    ],
+)
+
+objc_library(
+    name = "TFLBertQuestionAnswererObjcTestLibrary",
+    testonly = 1,
+    srcs = glob(["Tests/*.m"]),
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:albert_model",
+        "//tensorflow_lite_support/cc/test/testdata/task/text:mobile_bert_model",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLBertQuestionAnswerer",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLBertQuestionAnswererObjcTest",
+    size = "large",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLBertQuestionAnswererObjcTestLibrary",
+    ],
+)
diff --git a/tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.h b/tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.h
new file mode 100644
index 0000000..57b7c69
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.h
@@ -0,0 +1,74 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import <Foundation/Foundation.h>
+
+NS_ASSUME_NONNULL_BEGIN
+/**
+ * Struct to represent the logit and offset of the answer related to context.
+ */
+struct TFLPos {
+  int start;
+  int end;
+  float logit;
+};
+
+/**
+ * Class for the Answer to BertQuestionAnswerer.
+ */
+@interface TFLQAAnswer : NSObject
+@property(nonatomic) struct TFLPos pos;
+@property(nonatomic) NSString* text;
+@end
+
+/**
+ * BertQA task API, performs tokenization for models (BERT, Albert, etc.) in
+ * preprocess and returns most possible answers.
+ *
+ * In particular, the branch of BERT models use WordPiece tokenizer, and the
+ * branch of Albert models use SentencePiece tokenizer, respectively.
+ */
+@interface TFLBertQuestionAnswerer : NSObject
+
+/**
+ * Creates a BertQuestionAnswerer instance with an albert model or mobilebert
+ * model. The API expects a Bert based TFLite model with metadata containing
+ * the following information:
+ *  input_process_units: for Wordpiece/Sentencepiece Tokenizer
+ *  3 input tensors with names "ids", "mask" and "segment_ids"
+ *  2 output tensors with names "end_logits" and "start_logits"
+ * Sample models:
+ *   https://tfhub.dev/tensorflow/lite-model/albert_lite_base/squadv1/1
+ *   https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1
+ * @param modelPath The file path to the tflite model.
+ * @return A BertQuestionAnswerer instance.
+ */
++ (instancetype)questionAnswererWithModelPath:(NSString *)modelPath
+    NS_SWIFT_NAME(questionAnswerer(modelPath:));
+
+/**
+ * Answers question based on the context. Could be empty if no answer was found
+ * from the given context.
+ * 
+ * @param context Context the question bases on.
+ * @param question Question to ask.
+ *
+ * @return A list of answers to the question, reversely sorted by the
+ * probability of each answer.
+ */
+- (NSArray<TFLQAAnswer*>*)answerWithContext:(NSString*)context
+                                   question:(NSString*)question
+    NS_SWIFT_NAME(answer(context:question:));
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.m b/tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.m
new file mode 100644
index 0000000..fc1bd08
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.m
@@ -0,0 +1,71 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.h"
+#import "GTMDefines.h"
+#include "tensorflow_lite_support/cc/task/text/qa/bert_qa_c_api.h"
+
+NS_ASSUME_NONNULL_BEGIN
+
+@implementation TFLQAAnswer
+@synthesize pos;
+@synthesize text;
+@end
+
+@interface TFLBertQuestionAnswerer()
+/** BertQuestionAnswerer backed by C API */
+@property(nonatomic) BertQuestionAnswerer *bertQuestionAnswerer;
+@end
+
+@implementation TFLBertQuestionAnswerer
+
+- (void)dealloc {
+  BertQuestionAnswererDelete(_bertQuestionAnswerer);
+}
+
++ (instancetype)questionAnswererWithModelPath:(NSString *)modelPath {
+  BertQuestionAnswerer* bert_qa = BertQuestionAnswererFromFile(modelPath.UTF8String);
+
+  _GTMDevAssert(bert_qa, @"Failed to create BertQuestionAnswerer");
+  return [[TFLBertQuestionAnswerer alloc]
+      initWithBertQuestionAnswerer:bert_qa];
+}
+
+- (instancetype)initWithBertQuestionAnswerer:(BertQuestionAnswerer *)bertQuestionAnswerer {
+  self = [super init];
+  if (self) {
+    _bertQuestionAnswerer = bertQuestionAnswerer;
+  }
+  return self;
+}
+
+- (NSArray<TFLQAAnswer *> *)answerWithContext:(NSString *)context question:(NSString *)question {
+  struct QaAnswers *cAnswers =
+      BertQuestionAnswererAnswer(_bertQuestionAnswerer, context.UTF8String, question.UTF8String);
+  NSMutableArray<TFLQAAnswer *> *ret = [NSMutableArray arrayWithCapacity:cAnswers->size];
+  for (int i = 0; i < cAnswers->size; i++) {
+    struct QaAnswer cAnswer = cAnswers->answers[i];
+    TFLQAAnswer *answer = [[TFLQAAnswer alloc] init];
+    struct TFLPos pos = {.start = cAnswer.start,
+                         .end = cAnswer.end,
+                         .logit = cAnswer.logit};
+    [answer setPos:pos];
+    [answer setText:[NSString stringWithUTF8String:cAnswer.text]];
+    [ret addObject:answer];
+  }
+  BertQuestionAnswererQaAnswersDelete(cAnswers);
+  return ret;
+}
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/task/text/qa/Tests/TFLBertQuestionAnswererTest.m b/tensorflow_lite_support/ios/task/text/qa/Tests/TFLBertQuestionAnswererTest.m
new file mode 100644
index 0000000..9061063
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/qa/Tests/TFLBertQuestionAnswererTest.m
@@ -0,0 +1,72 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "tensorflow_lite_support/ios/task/text/qa/Sources/TFLBertQuestionAnswerer.h"
+
+#import <XCTest/XCTest.h>
+
+static NSString *const kContext =
+    @"The role of teacher is often formal and ongoing, carried out at a school "
+     "or other place of formal education. In many countries, a person who "
+     "wishes to become a teacher must first obtain specified professional "
+     "qualifications or credentials from a university or college. These "
+     "professional qualifications may include the study of pedagogy, the "
+     "science of teaching. Teachers, like other professionals, may have to "
+     "continue their education after they qualify, a process known as "
+     "continuing professional development. Teachers may use a lesson plan to "
+     "facilitate student learning, providing a course of study which is called "
+     "the curriculum.";
+static NSString *const kQuestion = @"What is a course of study called?";
+static NSString *const kAnswer = @"the curriculum.";
+
+@interface TFLBertQuestionAnswererTest : XCTestCase
+@property(nonatomic, nullable) NSString *mobileBertModelPath;
+@property(nonatomic, nullable) NSString *albertModelPath;
+@end
+
+@implementation TFLBertQuestionAnswererTest
+#pragma mark - Tests
+
+- (void)setUp {
+  [super setUp];
+  NSBundle *bundle = [NSBundle bundleForClass:[self class]];
+  self.mobileBertModelPath = [bundle pathForResource:@"mobilebert_with_metadata" ofType:@"tflite"];
+  self.albertModelPath = [bundle pathForResource:@"albert_with_metadata" ofType:@"tflite"];
+}
+
+- (void)testInitMobileBert {
+  TFLBertQuestionAnswerer* mobileBertAnswerer =
+    [TFLBertQuestionAnswerer questionAnswererWithModelPath:self.mobileBertModelPath];
+
+  XCTAssertNotNil(mobileBertAnswerer);
+
+  NSArray<TFLQAAnswer*>* answers =
+    [mobileBertAnswerer answerWithContext:kContext question:kQuestion];
+
+  XCTAssertEqualObjects([answers[0] text], kAnswer);
+}
+
+- (void)testInitAlbert {
+  TFLBertQuestionAnswerer* albertAnswerer =
+    [TFLBertQuestionAnswerer questionAnswererWithModelPath:self.albertModelPath];
+
+  XCTAssertNotNil(albertAnswerer);
+
+  NSArray<TFLQAAnswer*>* answers =
+    [albertAnswerer answerWithContext:kContext question:kQuestion];
+
+
+  XCTAssertEqualObjects([answers[0] text], kAnswer);
+}
+@end
diff --git a/tensorflow_lite_support/ios/task/text/qa/Tests/TFLBertQuestionAnswererTest.swift b/tensorflow_lite_support/ios/task/text/qa/Tests/TFLBertQuestionAnswererTest.swift
new file mode 100644
index 0000000..3f15cc5
--- /dev/null
+++ b/tensorflow_lite_support/ios/task/text/qa/Tests/TFLBertQuestionAnswererTest.swift
@@ -0,0 +1,63 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+import XCTest
+
+@testable import TFLBertQuestionAnswerer
+
+class TFLBertQuestionAnswererTest: XCTestCase {
+
+  static let bundle = Bundle(for: TFLBertQuestionAnswererTest.self)
+  static let mobileBertModelPath = bundle.path(forResource: "mobilebert_with_metadata", ofType: "tflite")!
+
+  static let albertModelPath = bundle.path(forResource: "albert_with_metadata", ofType: "tflite")!
+
+  static let context = """
+    The role of teacher is often formal and ongoing, carried out at a school or other place of
+    formal education. In many countries, a person who wishes to become a teacher must first obtain
+     specified professional qualifications or credentials from a university or college. These
+    professional qualifications may include the study of pedagogy, the science of teaching.
+    Teachers, like other professionals, may have to continue their education after they qualify,
+    a process known as continuing professional development. Teachers may use a lesson plan to
+    facilitate student learning, providing a course of study which is called the curriculum.
+    """
+  static let question = "What is a course of study called?"
+  static let answer = "the curriculum."
+
+  func testInitMobileBert() {
+    let mobileBertAnswerer = TFLBertQuestionAnswerer.questionAnswerer(
+      modelPath: TFLBertQuestionAnswererTest.mobileBertModelPath)
+
+    XCTAssertNotNil(mobileBertAnswerer)
+
+    let answers = mobileBertAnswerer.answer(
+      context: TFLBertQuestionAnswererTest.context, question: TFLBertQuestionAnswererTest.question)
+
+    XCTAssertNotNil(answers)
+    XCTAssertEqual(answers[0].text, TFLBertQuestionAnswererTest.answer)
+  }
+
+  func testInitAlbert() {
+    let albertAnswerer = TFLBertQuestionAnswerer.questionAnswerer(
+      modelPath: TFLBertQuestionAnswererTest.albertModelPath)
+
+    XCTAssertNotNil(albertAnswerer)
+
+    let answers = albertAnswerer.answer(
+      context: TFLBertQuestionAnswererTest.context, question: TFLBertQuestionAnswererTest.question)
+
+    XCTAssertNotNil(answers)
+    XCTAssertEqual(answers[0].text, TFLBertQuestionAnswererTest.answer)
+  }
+}
diff --git a/tensorflow_lite_support/ios/text/tokenizers/BUILD b/tensorflow_lite_support/ios/text/tokenizers/BUILD
new file mode 100644
index 0000000..34ba9c6
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/BUILD
@@ -0,0 +1,106 @@
+load("@build_bazel_rules_swift//swift:swift.bzl", "swift_library")
+load("@org_tensorflow//tensorflow/lite/experimental/ios:ios.bzl", "TFL_DEFAULT_TAGS", "TFL_DISABLED_SANITIZER_TAGS", "TFL_MINIMUM_OS_VERSION")
+load("@build_bazel_rules_apple//apple:ios.bzl", "ios_unit_test")
+load("@org_tensorflow//tensorflow/lite:special_rules.bzl", "tflite_ios_lab_runner")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+objc_library(
+    name = "TFLTokenizerUtil",
+    srcs = [
+        "Sources/TFLTokenizerUtil.mm",
+    ],
+    hdrs = [
+        "Sources/TFLTokenizerUtil.h",
+    ],
+    module_name = "TFLTokenizerUtil",
+    deps = [
+        "//tensorflow_lite_support/cc/text/tokenizers:tokenizer",
+        "//tensorflow_lite_support/ios/utils:TFLStringUtil",
+    ],
+)
+
+objc_library(
+    name = "TFLBertTokenizer",
+    srcs = [
+        "Sources/TFLBertTokenizer.mm",
+    ],
+    hdrs = [
+        "Sources/TFLBertTokenizer.h",
+        "Sources/TFLTokenizer.h",
+    ],
+    module_name = "TFLBertTokenizer",
+    deps = [
+        ":TFLTokenizerUtil",
+        "//tensorflow_lite_support/cc/text/tokenizers:bert_tokenizer",
+        "//tensorflow_lite_support/ios/utils:TFLStringUtil",
+    ],
+)
+
+swift_library(
+    name = "TFLBertTokenizerTestLibrary",
+    testonly = 1,
+    srcs = ["Tests/TFLBertTokenizerTest.swift"],
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:mobile_bert_model",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLBertTokenizer",
+        "//third_party/swift/xctest",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLBertTokenizerTest",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLBertTokenizerTestLibrary",
+    ],
+)
+
+objc_library(
+    name = "TFLSentencepieceTokenizer",
+    srcs = [
+        "Sources/TFLSentencepieceTokenizer.mm",
+    ],
+    hdrs = [
+        "Sources/TFLSentencepieceTokenizer.h",
+        "Sources/TFLTokenizer.h",
+    ],
+    module_name = "TFLSentencepieceTokenizer",
+    deps = [
+        ":TFLTokenizerUtil",
+        "//tensorflow_lite_support/cc/text/tokenizers:sentencepiece_tokenizer",
+        "//tensorflow_lite_support/ios/utils:TFLStringUtil",
+    ],
+)
+
+swift_library(
+    name = "TFLSentencepieceTokenizerTestLibrary",
+    testonly = 1,
+    srcs = ["Tests/TFLSentencepieceTokenizerTest.swift"],
+    data = [
+        "//tensorflow_lite_support/cc/test/testdata/task/text:albert_model",
+    ],
+    tags = TFL_DEFAULT_TAGS,
+    deps = [
+        ":TFLSentencepieceTokenizer",
+        "//third_party/swift/xctest",
+    ],
+)
+
+ios_unit_test(
+    name = "TFLSentencepieceTokenizerTest",
+    minimum_os_version = TFL_MINIMUM_OS_VERSION,
+    runner = tflite_ios_lab_runner("IOS_LATEST"),
+    tags = TFL_DEFAULT_TAGS + TFL_DISABLED_SANITIZER_TAGS,
+    deps = [
+        ":TFLSentencepieceTokenizerTestLibrary",
+    ],
+)
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLBertTokenizer.h b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLBertTokenizer.h
new file mode 100644
index 0000000..aa69248
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLBertTokenizer.h
@@ -0,0 +1,38 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "third_party/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizer.h"
+
+NS_ASSUME_NONNULL_BEGIN
+/**
+ * Wordpiece Tokenizer implemenation.
+ */
+@interface TFLBertTokenizer : NSObject <TFLTokenizer>
+
+/**
+ * Default initializer is not available.
+ */
+- (instancetype)init NS_UNAVAILABLE;
+
+/**
+ * Initializes the tokenizer with the path to wordpiece vocabulary file.
+ */
+- (instancetype)initWithVocabPath:(NSString *)vocabPath NS_DESIGNATED_INITIALIZER;
+
+/**
+ * Initializes the tokenizer with a list of tokens.
+ */
+- (instancetype)initWithVocab:(NSArray<NSString *> *)vocab NS_DESIGNATED_INITIALIZER;
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLBertTokenizer.mm b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLBertTokenizer.mm
new file mode 100644
index 0000000..949cef2
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLBertTokenizer.mm
@@ -0,0 +1,57 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "third_party/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLBertTokenizer.h"
+#include "third_party/tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h"
+#import "third_party/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.h"
+#import "third_party/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.h"
+
+NS_ASSUME_NONNULL_BEGIN
+using BertTokenizerCPP = ::tflite::support::text::tokenizer::BertTokenizer;
+
+@implementation TFLBertTokenizer {
+  std::unique_ptr<BertTokenizerCPP> _bertTokenizer;
+}
+
+- (instancetype)initWithVocabPath:(NSString *)vocabPath {
+  self = [super init];
+  if (self) {
+    _bertTokenizer = absl::make_unique<BertTokenizerCPP>(MakeString(vocabPath));
+  }
+  return self;
+}
+
+- (instancetype)initWithVocab:(NSArray<NSString *> *)vocab {
+  self = [super init];
+  if (self) {
+    std::vector<std::string> vocabCpp;
+    vocabCpp.reserve([vocab count]);
+    for (NSString *word in vocab) {
+      vocabCpp.emplace_back(MakeString(word));
+    }
+    _bertTokenizer = absl::make_unique<BertTokenizerCPP>(vocabCpp);
+  }
+  return self;
+}
+
+- (NSArray<NSString *> *)tokensFromInput:(NSString *)input {
+  return Tokenize(_bertTokenizer.get(), input);
+}
+
+- (NSArray<NSNumber *> *)idsFromTokens:(NSArray<NSString *> *)tokens {
+  return ConvertTokensToIds(_bertTokenizer.get(), tokens);
+}
+
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLSentencepieceTokenizer.h b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLSentencepieceTokenizer.h
new file mode 100644
index 0000000..eef3bf1
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLSentencepieceTokenizer.h
@@ -0,0 +1,33 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "third_party/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizer.h"
+
+NS_ASSUME_NONNULL_BEGIN
+/**
+ * Sentencepiece Tokenizer implemenation.
+ */
+@interface TFLSentencepieceTokenizer : NSObject <TFLTokenizer>
+
+/**
+ * Default initializer is not available.
+ */
+- (instancetype)init NS_UNAVAILABLE;
+
+/**
+ * Initializes the tokenizer with the path to sentencepiece model file.
+ */
+- (instancetype)initWithModelPath:(NSString *)modelPath;
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLSentencepieceTokenizer.mm b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLSentencepieceTokenizer.mm
new file mode 100644
index 0000000..1e21cee
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLSentencepieceTokenizer.mm
@@ -0,0 +1,45 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "third_party/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLSentencepieceTokenizer.h"
+#include "third_party/absl/memory/memory.h"
+#include "third_party/tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h"
+#import "third_party/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.h"
+#import "third_party/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.h"
+
+NS_ASSUME_NONNULL_BEGIN
+using SentencepieceTokenizerCPP = ::tflite::support::text::tokenizer::SentencePieceTokenizer;
+
+@implementation TFLSentencepieceTokenizer {
+  std::unique_ptr<SentencepieceTokenizerCPP> _spTokenizer;
+}
+
+- (instancetype)initWithModelPath:(NSString *)modelPath {
+  self = [super init];
+  if (self) {
+    _spTokenizer = absl::make_unique<SentencepieceTokenizerCPP>(MakeString(modelPath));
+  }
+  return self;
+}
+
+- (NSArray<NSString *> *)tokensFromInput:(NSString *)input {
+  return Tokenize(_spTokenizer.get(), input);
+}
+
+- (NSArray<NSNumber *> *)idsFromTokens:(NSArray<NSString *> *)tokens {
+  return ConvertTokensToIds(_spTokenizer.get(), tokens);
+}
+
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizer.h b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizer.h
new file mode 100644
index 0000000..ee0972f
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizer.h
@@ -0,0 +1,39 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import <Foundation/Foundation.h>
+
+NS_ASSUME_NONNULL_BEGIN
+/**
+ * Protocol for a Tokenizer used in model proprocessing.
+ */
+@protocol TFLTokenizer
+
+/**
+ * Performs tokenization on input text.
+ * @param input The input string to be tokenized.
+ *
+ * @return A list of tokens.
+ */
+- (NSArray<NSString *> *)tokensFromInput:(NSString *)input;
+
+/*
+ * Convert a list of tokens back to their coressponding IDs.
+ * @param tokens The tokens to be converted.
+ *
+ * @return A list of ids.
+ */
+- (NSArray<NSNumber *> *)idsFromTokens:(NSArray<NSString *> *)tokens;
+@end
+NS_ASSUME_NONNULL_END
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.h b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.h
new file mode 100644
index 0000000..574b555
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.h
@@ -0,0 +1,38 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import <Foundation/Foundation.h>
+#include "tensorflow_lite_support/cc/text/tokenizers/tokenizer.h"
+
+using ::tflite::support::text::tokenizer::Tokenizer;
+
+/**
+ * Invokes the cpp tokenizer's tokenize function and converts input/output to objc.
+ *
+ * @param tokenizer The cpp tokenizer pointer.
+ * @param input The input string to be tokenized.
+ *
+ * @return A list of tokens.
+ */
+NSArray<NSString *> *Tokenize(Tokenizer *tokenizer, NSString *input);
+
+/**
+ * Invokes the cpp tokenizer's convertTokensToIds function and converts input/output to objc.
+ *
+ * @param tokenizer The cpp tokenizer pointer.
+ * @param input The tokens to be converted.
+ *
+ * @return A list of ids.
+ */
+NSArray<NSNumber *> *ConvertTokensToIds(Tokenizer *tokenizer, NSArray<NSString *> *tokens);
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.mm b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.mm
new file mode 100644
index 0000000..5218057
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.mm
@@ -0,0 +1,41 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "third_party/tensorflow_lite_support/ios/text/tokenizers/Sources/TFLTokenizerUtil.h"
+
+#import "third_party/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.h"
+
+using ::tflite::support::text::tokenizer::TokenizerResult;
+
+NSArray<NSString *> *Tokenize(Tokenizer *tokenizer, NSString *input) {
+  TokenizerResult tokenize_result = tokenizer->Tokenize(MakeString(input));
+  std::vector<std::string> subwords = tokenize_result.subwords;
+  NSMutableArray<NSString *> *ret = [NSMutableArray arrayWithCapacity:subwords.size()];
+  for (int i = 0; i < subwords.size(); ++i) {
+    [ret addObject:MakeNSString(subwords[i])];
+  }
+  return ret;
+}
+
+NSArray<NSNumber *> *ConvertTokensToIds(Tokenizer *tokenizer, NSArray<NSString *> *tokens) {
+  NSMutableArray<NSNumber *> *ret = [NSMutableArray arrayWithCapacity:[tokens count]];
+  for (NSString *token in tokens) {
+    std::string cc_token = MakeString(token);
+    const char *cToken = cc_token.c_str();
+    int id;
+    tokenizer->LookupId(cToken, &id);
+    [ret addObject:[NSNumber numberWithInt:id]];
+  }
+  return ret;
+}
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Tests/TFLBertTokenizerTest.swift b/tensorflow_lite_support/ios/text/tokenizers/Tests/TFLBertTokenizerTest.swift
new file mode 100644
index 0000000..e805f30
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Tests/TFLBertTokenizerTest.swift
@@ -0,0 +1,50 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+import XCTest
+
+@testable import TFLBertTokenizer
+
+class TFLBertTokenizerTest: XCTestCase {
+  static let bundle = Bundle(for: TFLBertTokenizerTest.self)
+  static let mobileBertVocabPath = bundle.path(forResource: "vocab", ofType: "txt")!
+
+  func testInitBertTokenizerFromPath() {
+    let bertTokenizer = TFLBertTokenizer(vocabPath: TFLBertTokenizerTest.mobileBertVocabPath)
+
+    XCTAssertNotNil(bertTokenizer)
+
+    let tokens = bertTokenizer.tokens(fromInput: "i'm questionansweraskask")
+
+    XCTAssertEqual(tokens, ["i", "'", "m", "question", "##ans", "##wer", "##ask", "##ask"])
+
+    let ids = bertTokenizer.ids(fromTokens: tokens)
+
+    XCTAssertEqual(ids, [1045, 1005, 1049, 3160, 6962, 13777, 19895, 19895])
+  }
+
+  func testInitBertTokenizerFromVocab() {
+    let bertTokenizer = TFLBertTokenizer(vocab: ["hell", "##o", "wor", "##ld", "there"])
+
+    XCTAssertNotNil(bertTokenizer)
+
+    let tokens = bertTokenizer.tokens(fromInput: "hello there hello world")
+
+    XCTAssertEqual(tokens, ["hell", "##o", "there", "hell", "##o", "wor", "##ld"])
+
+    let ids = bertTokenizer.ids(fromTokens: tokens)
+
+    XCTAssertEqual(ids, [0, 1, 4, 0, 1, 2, 3])
+  }
+}
diff --git a/tensorflow_lite_support/ios/text/tokenizers/Tests/TFLSentencepieceTokenizerTest.swift b/tensorflow_lite_support/ios/text/tokenizers/Tests/TFLSentencepieceTokenizerTest.swift
new file mode 100644
index 0000000..c7c6d1e
--- /dev/null
+++ b/tensorflow_lite_support/ios/text/tokenizers/Tests/TFLSentencepieceTokenizerTest.swift
@@ -0,0 +1,37 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+import XCTest
+
+@testable import TFLSentencepieceTokenizer
+
+class TFLSentencepieceTokenizerTest: XCTestCase {
+  static let bundle = Bundle(for: TFLSentencepieceTokenizerTest.self)
+  static let spModelPath = bundle.path(forResource: "30k-clean", ofType: "model")!
+
+  func testInitSentenpieceTokenizerFromPath() {
+    let spTokenizer = TFLSentencepieceTokenizer(
+      modelPath: TFLSentencepieceTokenizerTest.spModelPath)
+
+    XCTAssertNotNil(spTokenizer)
+
+    let tokens = spTokenizer.tokens(fromInput: "good morning, i'm your teacher.\n")
+
+    XCTAssertEqual(tokens, ["▁good", "▁morning", ",", "▁i", "'", "m", "▁your", "▁teacher", "."])
+
+    let ids = spTokenizer.ids(fromTokens: tokens)
+
+    XCTAssertEqual(ids, [254, 959, 15, 31, 22, 79, 154, 2197, 9])
+  }
+}
diff --git a/tensorflow_lite_support/ios/utils/BUILD b/tensorflow_lite_support/ios/utils/BUILD
new file mode 100644
index 0000000..63f1091
--- /dev/null
+++ b/tensorflow_lite_support/ios/utils/BUILD
@@ -0,0 +1,15 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+objc_library(
+    name = "TFLStringUtil",
+    srcs = [
+        "Sources/TFLStringUtil.mm",
+    ],
+    hdrs = [
+        "Sources/TFLStringUtil.h",
+    ],
+    module_name = "TFLStringUtil",
+)
diff --git a/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.h b/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.h
new file mode 100644
index 0000000..3ea091f
--- /dev/null
+++ b/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.h
@@ -0,0 +1,23 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import <Foundation/Foundation.h>
+
+#include <string>
+
+// Translates a NSString encoded in UTF-8 to a std::string.
+std::string MakeString(NSString*);
+
+// Translates a std::string to the equivalent NSString by making a copy.
+NSString* MakeNSString(const std::string&);
diff --git a/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.mm b/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.mm
new file mode 100644
index 0000000..6e9cf23
--- /dev/null
+++ b/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.mm
@@ -0,0 +1,23 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#import "third_party/tensorflow_lite_support/ios/utils/Sources/TFLStringUtil.h"
+
+std::string MakeString(NSString* str) { return std::string([str UTF8String]); }
+
+NSString* MakeNSString(const std::string& str) {
+  return [[NSString alloc] initWithBytes:const_cast<void*>(static_cast<const void*>(str.data()))
+                                  length:str.length()
+                                encoding:NSUTF8StringEncoding];
+}
diff --git a/tensorflow_lite_support/java/AndroidManifest.xml b/tensorflow_lite_support/java/AndroidManifest.xml
new file mode 100644
index 0000000..1490929
--- /dev/null
+++ b/tensorflow_lite_support/java/AndroidManifest.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="org.tensorflow.lite.support">
+  <uses-sdk android:minSdkVersion="19" android:targetSdkVersion="29"/>
+</manifest>
diff --git a/tensorflow_lite_support/java/BUILD b/tensorflow_lite_support/java/BUILD
new file mode 100644
index 0000000..090b1ad
--- /dev/null
+++ b/tensorflow_lite_support/java/BUILD
@@ -0,0 +1,78 @@
+# Description:
+# TensorFlow Lite Support API in Java.
+
+load("@org_tensorflow//tensorflow/java:build_defs.bzl", "JAVACOPTS")
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+
+package(
+    default_visibility = ["//visibility:public"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "AndroidManifest.xml",
+    "default_version_script.lds",
+    "debug_version_script.lds",
+])
+
+# Android Library target for TFLite Support Library. It depends on TensorFlow
+# Lite runtime (tensorflow/lite/java:tensorflowlite). If you don't want to
+# introduce the native library into dependencies, use
+# "tensorflowlite_support_java" instead, which depends on
+# tensorflow/lite/java:tensorflowlite_java.
+android_library(
+    name = "tensorflowlite_support",
+    srcs = glob(
+        ["src/java/org/tensorflow/lite/support/**/*.java"],
+    ),
+    javacopts = JAVACOPTS,
+    manifest = "AndroidManifest.xml",
+    deps = [
+        "@org_checkerframework_qual",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite",
+    ],
+)
+
+android_library(
+    name = "tensorflowlite_support_java",
+    srcs = glob(
+        ["src/java/org/tensorflow/lite/support/**/*.java"],
+    ),
+    javacopts = JAVACOPTS,
+    manifest = "AndroidManifest.xml",
+    deps = [
+        "@org_checkerframework_qual",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# TODO(b/156482505): Remove this target.
+alias(
+    name = "tensorflow-lite-support-nogpu",
+    actual = ":tensorflow-lite-support",
+)
+
+# This alias matches the associated .aar library name output style.
+alias(
+    name = "tensorflow-lite-support",
+    actual = ":tensorflowlite_support",
+)
+
+java_library(
+    name = "tensorflowlite_support_precondition_lib",
+    srcs = ["src/java/org/tensorflow/lite/support/common/SupportPreconditions.java"],
+    javacopts = JAVACOPTS,
+    deps = [
+        "@org_checkerframework_qual",
+    ],
+)
+
+android_library(
+    name = "tensorflowlite_support_precondition",
+    srcs = ["src/java/org/tensorflow/lite/support/common/SupportPreconditions.java"],
+    javacopts = JAVACOPTS,
+    manifest = "AndroidManifest.xml",
+    deps = [
+        "@org_checkerframework_qual",
+    ],
+)
diff --git a/tensorflow_lite_support/java/README.md b/tensorflow_lite_support/java/README.md
new file mode 100644
index 0000000..8d37bf8
--- /dev/null
+++ b/tensorflow_lite_support/java/README.md
@@ -0,0 +1,38 @@
+# TensorFlow Lite Support
+
+TensorFlow Lite Support contains a set of tools and libraries that help
+developing ML with TFLite for mobile apps. See the [documentation on
+tensorflow.org](https://www.tensorflow.org/lite/inference_with_metadata/overview)
+for more information about all the efforts under TensorFlow Lite Support.
+
+This directory contains the Java code for the TensorFlow Lite SupportLibrary
+and TensorFlow Lite Task Library.
+
+## TensorFlow Lite Android Support Library
+
+Mobile application developers typically interact with typed objects such as
+bitmaps or primitives such as integers. However, the TensorFlow Lite Interpreter
+that runs the on-device machine learning model uses tensors in the form of
+ByteBuffer, which can be difficult to debug and manipulate. The TensorFlow Lite
+Android Support Library is designed to help process the input and output of
+TensorFlow Lite models, and make the TensorFlow Lite interpreter easier to use.
+
+We welcome feedback from the community as we develop this support library,
+especially around:
+
+*   Use-cases we should support including data types and operations
+*   Ease of use - does the APIs make sense to the community
+
+See the [documentation](https://www.tensorflow.org/lite/inference_with_metadata/lite_support)
+for more instruction and examples.
+
+
+## TensorFlow Lite Android Task Library
+TensorFlow Lite Task Library provides optimized ready-to-use model interfaces
+for popular machine learning tasks, such as image classification, question and
+answer, etc. The model interfaces are specifically designed for each task to
+achieve the best performance and usability. Task Library works cross-platform
+and is supported on Java, C++, and Swift.
+
+See the [documentation](https://www.tensorflow.org/lite/inference_with_metadata/task_library/overview)
+for more instruction and examples.
diff --git a/tensorflow_lite_support/java/debug_version_script.lds b/tensorflow_lite_support/java/debug_version_script.lds
new file mode 100644
index 0000000..53553a4
--- /dev/null
+++ b/tensorflow_lite_support/java/debug_version_script.lds
@@ -0,0 +1,5 @@
+VERS_1.0 {
+  # Export everything for debug purpose.
+  global:
+    *;
+};
diff --git a/tensorflow_lite_support/java/default_version_script.lds b/tensorflow_lite_support/java/default_version_script.lds
new file mode 100644
index 0000000..46bbffe
--- /dev/null
+++ b/tensorflow_lite_support/java/default_version_script.lds
@@ -0,0 +1,12 @@
+VERS_1.0 {
+  # Export JNI and native C symbols.
+  global:
+    Java_*;
+    JNI_OnLoad;
+    JNI_OnUnload;
+    TfLite*;
+
+  # Hide everything else.
+  local:
+    *;
+};
diff --git a/tensorflow_lite_support/java/jni/BUILD b/tensorflow_lite_support/java/jni/BUILD
new file mode 100644
index 0000000..2c01b50
--- /dev/null
+++ b/tensorflow_lite_support/java/jni/BUILD
@@ -0,0 +1,48 @@
+package(default_visibility = ["//tensorflow_lite_support:__subpackages__"])
+
+licenses(["notice"])  # Apache 2.0
+
+# Helper target for exposing JNI headers across multiple platforms.
+cc_library(
+    name = "jni",
+    hdrs = select({
+        # The Android toolchain makes "jni.h" available in the include path.
+        # For non-Android toolchains, generate jni.h and jni_md.h.
+        "//tensorflow_lite_support:android": [],
+        "//conditions:default": [
+            ":jni.h",
+            ":jni_md.h",
+        ],
+    }),
+    includes = select({
+        "//tensorflow_lite_support:android": [],
+        "//conditions:default": ["."],
+    }),
+    visibility = ["//visibility:public"],
+)
+
+# Silly rules to make
+# #include <jni.h>
+# in the source headers work
+# (in combination with the "includes" attribute of the tf_cuda_library rule
+# above. Not needed when using the Android toolchain).
+#
+# Inspired from:
+# https://github.com/bazelbuild/bazel/blob/f99a0543f8d97339d32075c7176b79f35be84606/src/main/native/BUILD
+# but hopefully there is a simpler alternative to this.
+genrule(
+    name = "copy_jni_h",
+    srcs = ["@bazel_tools//tools/jdk:jni_header"],
+    outs = ["jni.h"],
+    cmd = "cp -f $< $@",
+)
+
+genrule(
+    name = "copy_jni_md_h",
+    srcs = select({
+        "//tensorflow_lite_support:macos": ["@bazel_tools//tools/jdk:jni_md_header-darwin"],
+        "//conditions:default": ["@bazel_tools//tools/jdk:jni_md_header-linux"],
+    }),
+    outs = ["jni_md.h"],
+    cmd = "cp -f $< $@",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/FileUtil.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/FileUtil.java
new file mode 100644
index 0000000..e83fd40
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/FileUtil.java
@@ -0,0 +1,184 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common;
+
+import android.content.Context;
+import android.content.res.AssetFileDescriptor;
+import java.io.BufferedReader;
+import java.io.FileInputStream;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.nio.ByteBuffer;
+import java.nio.MappedByteBuffer;
+import java.nio.channels.FileChannel;
+import java.nio.charset.Charset;
+import java.util.ArrayList;
+import java.util.List;
+import org.checkerframework.checker.nullness.qual.NonNull;
+
+/** File I/O utilities. */
+public class FileUtil {
+  private FileUtil() {}
+
+  /**
+   * Loads labels from the label file into a list of strings.
+   *
+   * <p>A legal label file is the plain text file whose contents are split into lines, and each line
+   * is an individual value. The file should be in assets of the context.
+   *
+   * @param context The context holds assets.
+   * @param filePath The path of the label file, relative with assets directory.
+   * @return a list of labels.
+   * @throws IOException if error occurs to open or read the file.
+   */
+  @NonNull
+  public static List<String> loadLabels(@NonNull Context context, @NonNull String filePath)
+      throws IOException {
+    return loadLabels(context, filePath, Charset.defaultCharset());
+  }
+
+  /**
+   * Loads labels from the label file into a list of strings.
+   *
+   * <p>A legal label file is the plain text file whose contents are split into lines, and each line
+   * is an individual value. The empty lines will be ignored. The file should be in assets of the
+   * context.
+   *
+   * @param context The context holds assets.
+   * @param filePath The path of the label file, relative with assets directory.
+   * @param cs {@code Charset} to use when decoding content of label file.
+   * @return a list of labels.
+   * @throws IOException if error occurs to open or read the file.
+   */
+  @NonNull
+  public static List<String> loadLabels(
+      @NonNull Context context, @NonNull String filePath, Charset cs) throws IOException {
+    SupportPreconditions.checkNotNull(context, "Context cannot be null.");
+    SupportPreconditions.checkNotNull(filePath, "File path cannot be null.");
+    try (InputStream inputStream = context.getAssets().open(filePath)) {
+      return loadLabels(inputStream, cs);
+    }
+  }
+
+  /**
+   * Loads labels from an input stream of an opened label file. See details for label files in
+   * {@link FileUtil#loadLabels(Context, String)}.
+   *
+   * @param inputStream the input stream of an opened label file.
+   * @return a list of labels.
+   * @throws IOException if error occurs to open or read the file.
+   */
+  @NonNull
+  public static List<String> loadLabels(@NonNull InputStream inputStream) throws IOException {
+    return loadLabels(inputStream, Charset.defaultCharset());
+  }
+
+  /**
+   * Loads labels from an input stream of an opened label file. See details for label files in
+   * {@link FileUtil#loadLabels(Context, String)}.
+   *
+   * @param inputStream the input stream of an opened label file.
+   * @param cs {@code Charset} to use when decoding content of label file.
+   * @return a list of labels.
+   * @throws IOException if error occurs to open or read the file.
+   */
+  @NonNull
+  public static List<String> loadLabels(@NonNull InputStream inputStream, Charset cs)
+      throws IOException {
+    List<String> labels = new ArrayList<>();
+    try (BufferedReader reader = new BufferedReader(new InputStreamReader(inputStream, cs))) {
+      String line;
+      while ((line = reader.readLine()) != null) {
+        if (line.trim().length() > 0) {
+          labels.add(line);
+        }
+      }
+      return labels;
+    }
+  }
+
+  /**
+   * Loads a vocabulary file (a single-column text file) into a list of strings.
+   *
+   * <p>A vocabulary file is a single-column plain text file whose contents are split into lines,
+   * and each line is an individual value. The file should be in assets of the context.
+   *
+   * @param context The context holds assets.
+   * @param filePath The path of the vocabulary file, relative with assets directory.
+   * @return a list of vocabulary words.
+   * @throws IOException if error occurs to open or read the file.
+   */
+  @NonNull
+  public static List<String> loadSingleColumnTextFile(
+      @NonNull Context context, @NonNull String filePath, Charset cs) throws IOException {
+    return loadLabels(context, filePath, cs);
+  }
+
+  /**
+   * Loads vocabulary from an input stream of an opened vocabulary file (which is a single-column
+   * text file). See details for vocabulary files in {@link FileUtil#loadVocabularyFile(Context,
+   * String)}.
+   *
+   * @param inputStream the input stream of an opened vocabulary file.
+   * @return a list of vocabulary words.
+   * @throws IOException if error occurs to open or read the file.
+   */
+  @NonNull
+  public static List<String> loadSingleColumnTextFile(@NonNull InputStream inputStream, Charset cs)
+      throws IOException {
+    return loadLabels(inputStream, cs);
+  }
+
+  /**
+   * Loads a file from the asset folder through memory mapping.
+   *
+   * @param context Application context to access assets.
+   * @param filePath Asset path of the file.
+   * @return the loaded memory mapped file.
+   * @throws IOException if an I/O error occurs when loading the tflite model.
+   */
+  @NonNull
+  public static MappedByteBuffer loadMappedFile(@NonNull Context context, @NonNull String filePath)
+      throws IOException {
+    SupportPreconditions.checkNotNull(context, "Context should not be null.");
+    SupportPreconditions.checkNotNull(filePath, "File path cannot be null.");
+    try (AssetFileDescriptor fileDescriptor = context.getAssets().openFd(filePath);
+    FileInputStream inputStream = new FileInputStream(fileDescriptor.getFileDescriptor())) {
+      FileChannel fileChannel = inputStream.getChannel();
+      long startOffset = fileDescriptor.getStartOffset();
+      long declaredLength = fileDescriptor.getDeclaredLength();
+      return fileChannel.map(FileChannel.MapMode.READ_ONLY, startOffset, declaredLength);
+    }
+  }
+
+  /**
+   * Loads a binary file from the asset folder.
+   *
+   * @param context Application context to access assets.
+   * @param filePath Asset path of the file.
+   * @return the byte array for the binary file.
+   * @throws IOException if an I/O error occurs when loading file.
+   */
+  @NonNull
+  public static byte[] loadByteFromFile(@NonNull Context context, @NonNull String filePath)
+      throws IOException {
+    ByteBuffer buffer = loadMappedFile(context, filePath);
+    byte[] byteArray = new byte[buffer.remaining()];
+    buffer.get(byteArray);
+    return byteArray;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/Operator.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/Operator.java
new file mode 100644
index 0000000..38dfe88
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/Operator.java
@@ -0,0 +1,31 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common;
+
+/**
+ * The common interface for classes that carries an "apply" method, which converts T to another one.
+ * @param <T> The class which Operator handles.
+ */
+public interface Operator<T> {
+
+  /**
+   * Applies an operation on a T object, returning a T object.
+   *
+   * <p>Note: The returned object could probably be the same one with given input, and given input
+   * could probably be changed.
+   */
+  T apply(T x);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/Processor.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/Processor.java
new file mode 100644
index 0000000..07d7e2b
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/Processor.java
@@ -0,0 +1,23 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common;
+
+/**
+ * Processes T object with prepared {@link Operator<T>}.
+ */
+public interface Processor<T> {
+  T process(T input);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/SequentialProcessor.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/SequentialProcessor.java
new file mode 100644
index 0000000..ff0c640
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/SequentialProcessor.java
@@ -0,0 +1,82 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common;
+
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import org.checkerframework.checker.nullness.qual.NonNull;
+
+/**
+ * A processor base class that chains a serial of {@link Operator<T>} and executes them.
+ *
+ * <p>Typically, users could use its subclasses, e.g. {@link
+ * org.tensorflow.lite.support.image.ImageProcessor} rather than directly use this one.
+ *
+ * @param <T> The type that the Operator is handling.
+ */
+public class SequentialProcessor<T> implements Processor<T> {
+
+  /** List of operators added to this {@link SequentialProcessor}. */
+  protected final List<Operator<T>> operatorList;
+  /**
+   * The {@link Map} between the operator name and the corresponding op indexes in {@code
+   * operatorList}. An operator may be added multiple times into this {@link SequentialProcessor}.
+   */
+  protected final Map<String, List<Integer>> operatorIndex;
+
+  protected SequentialProcessor(Builder<T> builder) {
+    operatorList = builder.operatorList;
+    operatorIndex = Collections.unmodifiableMap(builder.operatorIndex);
+  }
+
+  @Override
+  public T process(T x) {
+    for (Operator<T> op : operatorList) {
+      x = op.apply(x);
+    }
+    return x;
+  }
+
+  /** The inner builder class to build a Sequential Processor. */
+  protected static class Builder<T> {
+
+    private final List<Operator<T>> operatorList;
+    private final Map<String, List<Integer>> operatorIndex;
+
+    protected Builder() {
+      operatorList = new ArrayList<>();
+      operatorIndex = new HashMap<>();
+    }
+
+    public Builder<T> add(@NonNull Operator<T> op) {
+      SupportPreconditions.checkNotNull(op, "Adding null Op is illegal.");
+      operatorList.add(op);
+      String operatorName = op.getClass().getName();
+      if (!operatorIndex.containsKey(operatorName)) {
+        operatorIndex.put(operatorName, new ArrayList<Integer>());
+      }
+      operatorIndex.get(operatorName).add(operatorList.size() - 1);
+      return this;
+    }
+
+    public SequentialProcessor<T> build() {
+      return new SequentialProcessor<T>(this);
+    }
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/SupportPreconditions.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/SupportPreconditions.java
new file mode 100644
index 0000000..8620e13
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/SupportPreconditions.java
@@ -0,0 +1,184 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common;
+
+import org.checkerframework.checker.nullness.qual.Nullable;
+
+/** Static error checking util methods. */
+public final class SupportPreconditions {
+  /**
+   * Ensures that an object reference passed as a parameter to the calling method is not null.
+   *
+   * @param reference an object reference
+   * @return the non-null reference that was validated
+   * @throws NullPointerException if {@code reference} is null
+   */
+  public static <T extends Object> T checkNotNull(T reference) {
+    if (reference == null) {
+      throw new NullPointerException("The object reference is null.");
+    }
+    return reference;
+  }
+
+  /**
+   * Ensures that an object reference passed as a parameter to the calling method is not null.
+   *
+   * @param reference an object reference
+   * @param errorMessage the exception message to use if the check fails; will be converted to a
+   *     string using {@link String#valueOf(Object)}
+   * @return the non-null reference that was validated
+   * @throws NullPointerException if {@code reference} is null
+   */
+  public static <T extends Object> T checkNotNull(T reference, @Nullable Object errorMessage) {
+    if (reference == null) {
+      throw new NullPointerException(String.valueOf(errorMessage));
+    }
+    return reference;
+  }
+
+  /**
+   * Ensures that the given String is not empty and not null.
+   *
+   * @param string the String to test
+   * @return the non-null non-empty String that was validated
+   * @throws IllegalArgumentException if {@code string} is null or empty
+   */
+  public static String checkNotEmpty(String string) {
+    if (string == null || string.length() == 0) {
+      throw new IllegalArgumentException("Given String is empty or null.");
+    }
+    return string;
+  }
+
+  /**
+   * Ensures that the given String is not empty and not null.
+   *
+   * @param string the String to test
+   * @param errorMessage the exception message to use if the check fails; will be converted to a
+   *     string using {@link String#valueOf(Object)}
+   * @return the non-null non-empty String that was validated
+   * @throws IllegalArgumentException if {@code string} is null or empty
+   */
+  public static String checkNotEmpty(String string, Object errorMessage) {
+    if (string == null || string.length() == 0) {
+      throw new IllegalArgumentException(String.valueOf(errorMessage));
+    }
+    return string;
+  }
+
+  /**
+   * Ensures the truth of an expression involving one or more parameters to the calling method.
+   *
+   * @param expression a boolean expression.
+   * @throws IllegalArgumentException if {@code expression} is false.
+   */
+  public static void checkArgument(boolean expression) {
+    if (!expression) {
+      throw new IllegalArgumentException();
+    }
+  }
+
+  /**
+   * Ensures the truth of an expression involving one or more parameters to the calling method.
+   *
+   * @param expression a boolean expression.
+   * @param errorMessage the exception message to use if the check fails; will be converted to a
+   *     string using {@link String#valueOf(Object)}.
+   * @throws IllegalArgumentException if {@code expression} is false.
+   */
+  public static void checkArgument(boolean expression, @Nullable Object errorMessage) {
+    if (!expression) {
+      throw new IllegalArgumentException(String.valueOf(errorMessage));
+    }
+  }
+
+  /**
+   * Ensures that {@code index} specifies a valid <i>element</i> in an array, list or string of size
+   * {@code size}. An element index may range from zero, inclusive, to {@code size}, exclusive.
+   *
+   * @param index a user-supplied index identifying an element of an array, list or string
+   * @param size the size of that array, list or string
+   * @return the value of {@code index}
+   * @throws IndexOutOfBoundsException if {@code index} is negative or is not less than {@code size}
+   * @throws IllegalArgumentException if {@code size} is negative
+   */
+  public static int checkElementIndex(int index, int size) {
+    return checkElementIndex(index, size, "index");
+  }
+
+  /**
+   * Ensures that {@code index} specifies a valid <i>element</i> in an array, list or string of size
+   * {@code size}. An element index may range from zero, inclusive, to {@code size}, exclusive.
+   *
+   * @param index a user-supplied index identifying an element of an array, list or string
+   * @param size the size of that array, list or string
+   * @param desc the text to use to describe this index in an error message
+   * @return the value of {@code index}
+   * @throws IndexOutOfBoundsException if {@code index} is negative or is not less than {@code size}
+   * @throws IllegalArgumentException if {@code size} is negative
+   */
+  public static int checkElementIndex(int index, int size, @Nullable String desc) {
+    // Carefully optimized for execution by hotspot (explanatory comment above)
+    if (index < 0 || index >= size) {
+      throw new IndexOutOfBoundsException(badElementIndex(index, size, desc));
+    }
+    return index;
+  }
+
+  /**
+   * Ensures the truth of an expression involving the state of the calling instance, but not
+   * involving any parameters to the calling method.
+   *
+   * @param expression a boolean expression
+   * @throws IllegalStateException if {@code expression} is false
+   * @see Verify#verify Verify.verify()
+   */
+  public static void checkState(boolean expression) {
+    if (!expression) {
+      throw new IllegalStateException();
+    }
+  }
+
+  /**
+   * Ensures the truth of an expression involving the state of the calling instance, but not
+   * involving any parameters to the calling method.
+   *
+   * @param expression a boolean expression
+   * @param errorMessage the exception message to use if the check fails; will be converted to a
+   *     string using {@link String#valueOf(Object)}
+   * @throws IllegalStateException if {@code expression} is false
+   * @see Verify#verify Verify.verify()
+   */
+  public static void checkState(boolean expression, @Nullable Object errorMessage) {
+    if (!expression) {
+      throw new IllegalStateException(String.valueOf(errorMessage));
+    }
+  }
+
+  private static String badElementIndex(int index, int size, @Nullable String desc) {
+    if (index < 0) {
+      return String.format("%s (%s) must not be negative", desc, index);
+    } else if (size < 0) {
+      throw new IllegalArgumentException("negative size: " + size);
+    } else { // index >= size
+      return String.format("%s (%s) must be less than size (%s)", desc, index, size);
+    }
+  }
+
+  private SupportPreconditions() {
+    throw new AssertionError("SupportPreconditions is Uninstantiable.");
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/TensorOperator.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/TensorOperator.java
new file mode 100644
index 0000000..d1b7021
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/TensorOperator.java
@@ -0,0 +1,27 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common;
+
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Applies some operation on TensorBuffers.
+ */
+public interface TensorOperator extends Operator<TensorBuffer> {
+  /** @see Operator#apply(Object) . */
+  @Override
+  TensorBuffer apply(TensorBuffer input);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/TensorProcessor.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/TensorProcessor.java
new file mode 100644
index 0000000..31531b2
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/TensorProcessor.java
@@ -0,0 +1,68 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common;
+
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * TensorProcessor is a helper class for preprocessing and postprocessing tensors. It could
+ * transform a {@link TensorBuffer} to another by executing a chain of {@link TensorOperator}.
+ *
+ * <p>Example Usage:
+ *
+ * <pre>
+ *   TensorProcessor processor = new TensorProcessor.Builder().add(new NormalizeOp(1, 2)).build();
+ *   TensorBuffer anotherTensorBuffer = processor.process(tensorBuffer);
+ * </pre>
+ *
+ * @see TensorProcessor.Builder to build a {@link TensorProcessor} instance.
+ * @see TensorProcessor#process(TensorBuffer) to apply the processor on a {@link TensorBuffer}.
+ */
+public class TensorProcessor extends SequentialProcessor<TensorBuffer> {
+  private TensorProcessor(Builder builder) {
+    super(builder);
+  }
+
+  /** The Builder to create an {@link TensorProcessor}, which could be executed later. */
+  public static class Builder extends SequentialProcessor.Builder<TensorBuffer> {
+
+    /**
+     * Creates a Builder to build {@link TensorProcessor}.
+     *
+     * @see #add(TensorOperator) to add an Op.
+     * @see #build() to complete the building process and get a built Processor.
+     */
+    public Builder() {
+      super();
+    }
+
+    /**
+     * Adds an {@link TensorOperator} into the Operator chain.
+     *
+     * @param op the Operator instance to be executed then.
+     */
+    public TensorProcessor.Builder add(TensorOperator op) {
+      super.add(op);
+      return this;
+    }
+
+    /** Completes the building process and gets the {@link TensorProcessor} instance. */
+    @Override
+    public TensorProcessor build() {
+      return new TensorProcessor(this);
+    }
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/CastOp.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/CastOp.java
new file mode 100644
index 0000000..3355b18
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/CastOp.java
@@ -0,0 +1,55 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common.ops;
+
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+import org.tensorflow.lite.support.common.TensorOperator;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/** Casts a {@link TensorBuffer} to a specified data type. */
+public class CastOp implements TensorOperator {
+
+  private final DataType destinationType;
+
+  /**
+   * Constructs a CastOp.
+   *
+   * <p>Note: For only converting type for a certain {@link TensorBuffer} on-the-fly rather than in
+   * a processor, please directly use {@link TensorBuffer#createFrom(TensorBuffer, DataType)}.
+   *
+   * <p>When this Op is executed, if the original {@link TensorBuffer} is already in {@code
+   * destinationType}, the original buffer will be directly returned.
+   *
+   * @param destinationType: The type of the casted {@link TensorBuffer}.
+   * @throws IllegalArgumentException if {@code destinationType} is neither {@link DataType#UINT8}
+   * nor {@link DataType#FLOAT32}.
+   */
+  public CastOp(DataType destinationType) {
+    SupportPreconditions.checkArgument(
+        destinationType == DataType.UINT8 || destinationType == DataType.FLOAT32,
+        "Destination type " + destinationType + " is not supported.");
+    this.destinationType = destinationType;
+  }
+
+  @Override
+  public TensorBuffer apply(TensorBuffer input) {
+    if (input.getDataType() == destinationType) {
+      return input;
+    }
+    return TensorBuffer.createFrom(input, destinationType);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/DequantizeOp.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/DequantizeOp.java
new file mode 100644
index 0000000..1881747
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/DequantizeOp.java
@@ -0,0 +1,40 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common.ops;
+
+import org.tensorflow.lite.support.common.TensorOperator;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Dequantizes a {@link TensorBuffer} with given {@code zeroPoint} and {@code scale}.
+ *
+ * <p>Note: The data type of output tensor is always {@code FLOAT32} except when the DequantizeOp is
+ * created effectively as an identity Op such as setting {@code zeroPoint} to 0 and {@code scale} to
+ * 1 (in this case, the output tensor is the same instance as input).
+ *
+ * <p>If both {@code zeroPoint} and {@code scale} are 0, the {@link DequantizeOp} will be bypassed,
+ * which is equivalent to setting {@code zeroPoint} to 0 and {@code scale} to 1. This can be useful
+ * when passing in the quantization parameters that are extracted directly from the TFLite model
+ * flatbuffer. If the tensor is not quantized, both {@code zeroPoint} and {@code scale} will be read
+ * as 0.
+ */
+public class DequantizeOp extends NormalizeOp implements TensorOperator {
+
+  public DequantizeOp(float zeroPoint, float scale) {
+    // Quantization: f = (q - z) * s
+    super(zeroPoint, 1 / scale);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/NormalizeOp.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/NormalizeOp.java
new file mode 100644
index 0000000..9db1388
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/NormalizeOp.java
@@ -0,0 +1,160 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common.ops;
+
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+import org.tensorflow.lite.support.common.TensorOperator;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+import org.tensorflow.lite.support.tensorbuffer.TensorBufferFloat;
+
+/**
+ * Normalizes a {@link TensorBuffer} with given mean and stddev: output = (input - mean) / stddev.
+ */
+public class NormalizeOp implements TensorOperator {
+
+  // mean.length should always be equal to stddev.length and always >= 1.
+  private final float[] mean;
+  private final float[] stddev;
+  private final int numChannels;
+  private final boolean isIdentityOp;
+
+  /**
+   * Initializes a NormalizeOp. When being called, it creates a new {@link TensorBuffer}, which
+   * satisfies:
+   *
+   * <pre>
+   *   output = (input - mean) / stddev
+   * </pre>
+   *
+   * <p>In the following two cases, reset {@code mean} to 0 and {@code stddev} to 1 to bypass the
+   * normalization. <br>
+   * 1. Both {@code mean} and {code stddev} are 0. <br>
+   * 2. {@code mean} is 0 and {stddev} is Infinity.
+   *
+   * <p>Note: If {@code mean} is set to 0 and {@code stddev} is set to 1, no computation will
+   * happen, and original input will be directly returned in execution.
+   *
+   * <p>Note: The returned {@link TensorBuffer} is always a {@link DataType#FLOAT32} tensor at
+   * present, except when the input is a {@link DataType#UINT8} tensor, {@code mean} is set to 0 and
+   * {@code stddev} is set to 1, so that the original {@link DataType#UINT8} tensor is returned.
+   *
+   * @param mean the mean value to be subtracted first.
+   * @param stddev the standard deviation value to divide then.
+   * @throws IllegalArgumentException if {@code stddev} is zero.
+   */
+  public NormalizeOp(float mean, float stddev) {
+    // Make exceptions to the cases that
+    // 1. Both mean and stddev are 0.0f. This may happen when reading the normalization parameters
+    // from a tensor which does not have the values populated in the metadata. The same situation
+    // may also happen to the quantization parameters.
+    // 2. mean is 0.0f and stddev is Infinity. This may happen when reading the quantization
+    // parameters from a tensor which does not have the values populated in the metadata, and then
+    // passing the parameters into the DequantizeOp.
+    // Bypass both of the two cases, by reseting stddev to 1.0f.
+    if (mean == 0.0f && (stddev == 0.0f || Float.isInfinite(stddev))) {
+      stddev = 1.0f;
+    }
+
+    SupportPreconditions.checkArgument(stddev != 0.0f, "Stddev cannot be zero.");
+    boolean meansIsZeroAndDevsIs1 = false;
+    if (mean == 0.0f && stddev == 1.0f) {
+      meansIsZeroAndDevsIs1 = true;
+    }
+
+    this.isIdentityOp = meansIsZeroAndDevsIs1;
+    this.mean = new float[] {mean};
+    this.stddev = new float[] {stddev};
+    this.numChannels = 1;
+  }
+
+  /**
+   * Initializes a NormalizeOp. When being called, it creates a new {@link TensorBuffer}, which
+   * satisfies:
+   *
+   * <pre>
+   *   // Pseudo code. [...][i] means a certain element whose channel id is i.
+   *   output[...][i] = (input[...][i] - mean[i]) / stddev[i]
+   * </pre>
+   *
+   * <p>Note: If all values in {@code mean} are set to 0 and all {@code stddev} are set to 1, no
+   * computation will happen, and original input will be directly returned in execution.
+   *
+   * <p>Note: The returned {@link TensorBuffer} is always a {@link DataType#FLOAT32} tensor at
+   * present, except that the input is a {@link DataType#UINT8} tensor, all {@code mean} are set to
+   * 0 and all {@code stddev} are set to 1.
+   *
+   * @param mean the mean values to be subtracted first for each channel.
+   * @param stddev the standard deviation values to divide then for each channel.
+   * @throws IllegalArgumentException if any {@code stddev} is zero, or {@code mean} has different
+   *     number of elements with {@code stddev}, or any of them is empty.
+   */
+  public NormalizeOp(@NonNull float[] mean, @NonNull float[] stddev) {
+    SupportPreconditions.checkNotNull(mean, "Mean cannot be null");
+    SupportPreconditions.checkNotNull(stddev, "Stddev cannot be null");
+    SupportPreconditions.checkArgument(
+        mean.length == stddev.length,
+        "Per channel normalization requires same number of means and stddevs");
+    SupportPreconditions.checkArgument(mean.length > 0, "Means and stddevs are empty.");
+    this.mean = mean.clone();
+    this.stddev = stddev.clone();
+    boolean allMeansAreZeroAndAllDevsAre1 = true;
+    this.numChannels = mean.length;
+    for (int i = 0; i < numChannels; i++) {
+      SupportPreconditions.checkArgument(this.stddev[i] != 0, "Stddev cannot be zero.");
+      if (this.stddev[i] != 1 || this.mean[i] != 0) {
+        allMeansAreZeroAndAllDevsAre1 = false;
+      }
+    }
+    this.isIdentityOp = allMeansAreZeroAndAllDevsAre1;
+  }
+
+  /**
+   * Applies the defined normalization on given tensor and returns the result.
+   *
+   * <p>Note: {@code input} is possibly the same instance with the output.
+   *
+   * @param input input tensor. It may be the same instance with the output.
+   * @return output tensor.
+   */
+  @Override
+  @NonNull
+  public TensorBuffer apply(@NonNull TensorBuffer input) {
+    if (isIdentityOp) {
+      return input;
+    }
+    int[] shape = input.getShape();
+    SupportPreconditions.checkArgument(
+        numChannels == 1 || (shape.length != 0 && shape[shape.length - 1] == numChannels),
+        "Number of means (stddevs) is not same with number of channels (size of last axis).");
+    // TODO(136750944): Eliminate the array copy here.
+    float[] values = input.getFloatArray();
+    int j = 0;
+    for (int i = 0; i < values.length; i++) {
+      values[i] = (values[i] - mean[j]) / stddev[j];
+      j = (j + 1) % numChannels;
+    }
+    TensorBuffer output;
+    if (input.isDynamic()) {
+      output = TensorBufferFloat.createDynamic(DataType.FLOAT32);
+    } else {
+      output = TensorBufferFloat.createFixedSize(shape, DataType.FLOAT32);
+    }
+    output.loadArray(values, shape);
+    return output;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/QuantizeOp.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/QuantizeOp.java
new file mode 100644
index 0000000..8b3e82a
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/common/ops/QuantizeOp.java
@@ -0,0 +1,41 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.common.ops;
+
+import org.tensorflow.lite.support.common.TensorOperator;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Quantizes a {@link TensorBuffer} with given {@code zeroPoint} and {@code scale}.
+ *
+ * <p>Note: {@link QuantizeOp} does not cast output to UINT8, but only performs the quantization
+ * math on top of input. The data type of output tensor is always {@code FLOAT32} except that the Op
+ * is effectively an identity Op (in this case, the output tensor is the same instance as the
+ * input). To connect with quantized model, a {@link CastOp} is probably needed.
+ *
+ * <p>If both {@code zeroPoint} and {@code scale} are 0, the {@link QuantizeOp} will be bypassed,
+ * which is equivalent to setting {@code zeroPoint} to 0 and {@code scale} to 1. This can be useful
+ * when passing in the quantization parameters that are extracted directly from the TFLite model
+ * flatbuffer. If the tensor is not quantized, both {@code zeroPoint} and {@code scale} will be read
+ * as 0.
+ */
+public class QuantizeOp extends NormalizeOp implements TensorOperator {
+
+  public QuantizeOp(float zeroPoint, float scale) {
+    // Quantization: f = (q - z) * s, i.e. q = f / s + z = (f - (-z * s)) / s
+    super(-zeroPoint * scale, scale);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/BitmapContainer.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/BitmapContainer.java
new file mode 100644
index 0000000..b9590bf
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/BitmapContainer.java
@@ -0,0 +1,80 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkArgument;
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkNotNull;
+
+import android.graphics.Bitmap;
+import android.graphics.Bitmap.Config;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/** Holds a {@link Bitmap} and converts it to other image formats as needed. */
+final class BitmapContainer implements ImageContainer {
+
+  private final Bitmap bitmap;
+
+  /**
+   * Creates a {@link BitmapContainer} object with ARGB_8888 {@link Bitmap}.
+   *
+   * @throws IllegalArgumentException if the bitmap configuration is not ARGB_8888
+   */
+  static BitmapContainer create(Bitmap bitmap) {
+    return new BitmapContainer(bitmap);
+  }
+
+  private BitmapContainer(Bitmap bitmap) {
+    checkNotNull(bitmap, "Cannot load null bitmap.");
+    checkArgument(
+        bitmap.getConfig().equals(Config.ARGB_8888), "Only supports loading ARGB_8888 bitmaps.");
+    this.bitmap = bitmap;
+  }
+
+  @Override
+  public BitmapContainer clone() {
+    return create(bitmap.copy(bitmap.getConfig(), bitmap.isMutable()));
+  }
+
+  @Override
+  public Bitmap getBitmap() {
+    // Not making a defensive copy for performance considerations. During image processing,
+    // users may need to set and get the bitmap many times.
+    return bitmap;
+  }
+
+  @Override
+  public TensorBuffer getTensorBuffer(DataType dataType) {
+    TensorBuffer buffer = TensorBuffer.createDynamic(dataType);
+    ImageConversions.convertBitmapToTensorBuffer(bitmap, buffer);
+    return buffer;
+  }
+
+  @Override
+  public int getWidth() {
+    return bitmap.getWidth();
+  }
+
+  @Override
+  public int getHeight() {
+    return bitmap.getHeight();
+  }
+
+  @Override
+  public ColorSpaceType getColorSpaceType() {
+    return ColorSpaceType.fromBitmapConfig(bitmap.getConfig());
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/BoundingBoxUtil.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/BoundingBoxUtil.java
new file mode 100644
index 0000000..1a46330
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/BoundingBoxUtil.java
@@ -0,0 +1,244 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkArgument;
+
+import android.graphics.RectF;
+import java.nio.ByteBuffer;
+import java.nio.FloatBuffer;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.List;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Helper class for converting values that represents bounding boxes into rectangles.
+ *
+ * <p>The class provides a static function to create bounding boxes as {@link RectF} from different
+ * types of configurations.
+ *
+ * <p>Generally, a bounding box could be represented by 4 float values, but the values could be
+ * interpreted in many ways. We now support 3 {@link Type} of configurations, and the order of
+ * elements in each type is configurable as well.
+ */
+public final class BoundingBoxUtil {
+
+  /** Denotes how a bounding box is represented. */
+  public enum Type {
+    /**
+     * Represents the bounding box by using the combination of boundaries, {left, top, right,
+     * bottom}. The default order is {left, top, right, bottom}. Other orders can be indicated by an
+     * index array.
+     */
+    BOUNDARIES,
+    /**
+     * Represents the bounding box by using the upper_left corner, width and height. The default
+     * order is {upper_left_x, upper_left_y, width, height}. Other orders can be indicated by an
+     * index array.
+     */
+    UPPER_LEFT,
+    /**
+     * Represents the bounding box by using the center of the box, width and height. The default
+     * order is {center_x, center_y, width, height}. Other orders can be indicated by an index
+     * array.
+     */
+    CENTER,
+  }
+
+  /** Denotes if the coordinates are actual pixels or relative ratios. */
+  public enum CoordinateType {
+    /** The coordinates are relative ratios in range [0, 1]. */
+    RATIO,
+    /** The coordinates are actual pixel values. */
+    PIXEL
+  }
+
+  /**
+   * Creates a list of bounding boxes from a {@link TensorBuffer} which represents bounding boxes.
+   *
+   * @param tensor holds the data representing some boxes.
+   * @param valueIndex denotes the order of the elements defined in each bounding box type. An empty
+   *     index array represent the default order of each bounding box type. For example, to denote
+   *     the default order of BOUNDARIES, {left, top, right, bottom}, the index should be {0, 1, 2,
+   *     3}. To denote the order {left, right, top, bottom}, the order should be {0, 2, 1, 3}.
+   *     <p>The index array can be applied to all bounding box types to adjust the order of their
+   *     corresponding underlying elements.
+   * @param boundingBoxAxis specifies the index of the dimension that represents bounding box. The
+   *     size of that dimension is required to be 4. Index here starts from 0. For example, if the
+   *     tensor has shape 4x10, the axis for bounding boxes is likely to be 0. Negative axis is also
+   *     supported: -1 gives the last axis and -2 gives the second, .etc. theFor shape 10x4, the
+   *     axis is likely to be 1 (or -1, equivalently).
+   * @param type defines how values should be converted into boxes. See {@link Type}
+   * @param coordinateType defines how values are interpreted to coordinates. See {@link
+   *     CoordinateType}
+   * @param height the height of the image which the boxes belong to. Only has effects when {@code
+   *     coordinateType} is {@link CoordinateType#RATIO}
+   * @param width the width of the image which the boxes belong to. Only has effects when {@code
+   *     coordinateType} is {@link CoordinateType#RATIO}
+   * @return A list of bounding boxes that the {@code tensor} represents. All dimensions except
+   *     {@code boundingBoxAxis} will be collapsed with order kept. For example, given {@code
+   *     tensor} with shape {1, 4, 10, 2} and {@code boundingBoxAxis = 1}, The result will be a list
+   *     of 20 bounding boxes.
+   * @throws IllegalArgumentException if size of bounding box dimension (set by {@code
+   *     boundingBoxAxis}) is not 4.
+   * @throws IllegalArgumentException if {@code boundingBoxAxis} is not in {@code (-(D+1), D)} where
+   *     {@code D} is the number of dimensions of the {@code tensor}.
+   * @throws IllegalArgumentException if {@code tensor} has data type other than {@link
+   *     DataType#FLOAT32}.
+   */
+  public static List<RectF> convert(
+      TensorBuffer tensor,
+      int[] valueIndex,
+      int boundingBoxAxis,
+      Type type,
+      CoordinateType coordinateType,
+      int height,
+      int width) {
+    int[] shape = tensor.getShape();
+    checkArgument(
+        boundingBoxAxis >= -shape.length && boundingBoxAxis < shape.length,
+        String.format(
+            "Axis %d is not in range (-(D+1), D), where D is the number of dimensions of input"
+                + " tensor (shape=%s)",
+            boundingBoxAxis, Arrays.toString(shape)));
+    if (boundingBoxAxis < 0) {
+      boundingBoxAxis = shape.length + boundingBoxAxis;
+    }
+    checkArgument(
+        shape[boundingBoxAxis] == 4,
+        String.format(
+            "Size of bounding box dimension %d is not 4. Got %d in shape %s",
+            boundingBoxAxis, shape[boundingBoxAxis], Arrays.toString(shape)));
+    checkArgument(
+        valueIndex.length == 4,
+        String.format(
+            "Bounding box index array length %d is not 4. Got index array %s",
+            valueIndex.length, Arrays.toString(valueIndex)));
+    checkArgument(
+        tensor.getDataType() == DataType.FLOAT32,
+        "Bounding Boxes only create from FLOAT32 buffers. Got: " + tensor.getDataType().name());
+    List<RectF> boundingBoxList = new ArrayList<>();
+    // Collapse dimensions to {a, 4, b}. So each bounding box could be represent as (i, j), and its
+    // four values are (i, k, j), where 0 <= k < 4. We can compute the 4 flattened index by
+    // i * 4b + k * b + j.
+    int a = 1;
+    for (int i = 0; i < boundingBoxAxis; i++) {
+      a *= shape[i];
+    }
+    int b = 1;
+    for (int i = boundingBoxAxis + 1; i < shape.length; i++) {
+      b *= shape[i];
+    }
+    float[] values = new float[4];
+    ByteBuffer byteBuffer = tensor.getBuffer();
+    byteBuffer.rewind();
+    FloatBuffer floatBuffer = byteBuffer.asFloatBuffer();
+    for (int i = 0; i < a; i++) {
+      for (int j = 0; j < b; j++) {
+        for (int k = 0; k < 4; k++) {
+          values[k] = floatBuffer.get((i * 4 + k) * b + j);
+        }
+        boundingBoxList.add(
+            convertOneBoundingBox(values, valueIndex, type, coordinateType, height, width));
+      }
+    }
+    byteBuffer.rewind();
+    return boundingBoxList;
+  }
+
+  private static RectF convertOneBoundingBox(
+      float[] values,
+      int[] valueIndex,
+      Type type,
+      CoordinateType coordinateType,
+      int height,
+      int width) {
+    float[] orderedValues = new float[4];
+    for (int i = 0; i < 4; i++) {
+      orderedValues[i] = values[valueIndex[i]];
+    }
+    return convertOneBoundingBox(orderedValues, type, coordinateType, height, width);
+  }
+
+  private static RectF convertOneBoundingBox(
+      float[] values, Type type, CoordinateType coordinateType, int height, int width) {
+    switch (type) {
+      case BOUNDARIES:
+        return convertFromBoundaries(values, coordinateType, height, width);
+      case UPPER_LEFT:
+        return convertFromUpperLeft(values, coordinateType, height, width);
+      case CENTER:
+        return convertFromCenter(values, coordinateType, height, width);
+    }
+    throw new IllegalArgumentException("Cannot recognize BoundingBox.Type " + type);
+  }
+
+  private static RectF convertFromBoundaries(
+      float[] values, CoordinateType coordinateType, int imageHeight, int imageWidth) {
+    float left = values[0];
+    float top = values[1];
+    float right = values[2];
+    float bottom = values[3];
+    return getRectF(left, top, right, bottom, imageHeight, imageWidth, coordinateType);
+  }
+
+  private static RectF convertFromUpperLeft(
+      float[] values, CoordinateType coordinateType, int imageHeight, int imageWidth) {
+    float left = values[0];
+    float top = values[1];
+    float right = values[0] + values[2];
+    float bottom = values[1] + values[3];
+    return getRectF(left, top, right, bottom, imageHeight, imageWidth, coordinateType);
+  }
+
+  private static RectF convertFromCenter(
+      float[] values, CoordinateType coordinateType, int imageHeight, int imageWidth) {
+    float centerX = values[0];
+    float centerY = values[1];
+    float w = values[2];
+    float h = values[3];
+
+    float left = centerX - w / 2;
+    float top = centerY - h / 2;
+    float right = centerX + w / 2;
+    float bottom = centerY + h / 2;
+    return getRectF(left, top, right, bottom, imageHeight, imageWidth, coordinateType);
+  }
+
+  private static RectF getRectF(
+      float left,
+      float top,
+      float right,
+      float bottom,
+      int imageHeight,
+      int imageWidth,
+      CoordinateType coordinateType) {
+    if (coordinateType == CoordinateType.PIXEL) {
+      return new RectF(
+          left, top, right, bottom);
+    } else if (coordinateType == CoordinateType.RATIO) {
+      return new RectF(
+          left * imageWidth, top * imageHeight, right * imageWidth, bottom * imageHeight);
+    } else {
+      throw new IllegalArgumentException("Cannot convert coordinate type " + coordinateType);
+    }
+  }
+
+  // Private constructor to prevent initialization.
+  private BoundingBoxUtil() {}
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ColorSpaceType.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ColorSpaceType.java
new file mode 100644
index 0000000..e92d095
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ColorSpaceType.java
@@ -0,0 +1,212 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkArgument;
+
+import android.graphics.Bitmap;
+import android.graphics.Bitmap.Config;
+import java.util.Arrays;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/** Represents the type of color space of an image. */
+public enum ColorSpaceType {
+  /** Each pixel has red, green, and blue color components. */
+  RGB {
+
+    // The channel axis should always be 3 for RGB images.
+    private static final int CHANNEL_VALUE = 3;
+
+    @Override
+    Bitmap convertTensorBufferToBitmap(TensorBuffer buffer) {
+      return ImageConversions.convertRgbTensorBufferToBitmap(buffer);
+    }
+
+    @Override
+    int getChannelValue() {
+      return CHANNEL_VALUE;
+    }
+
+    @Override
+    int[] getNormalizedShape(int[] shape) {
+      switch (shape.length) {
+          // The shape is in (h, w, c) format.
+        case 3:
+          return insertValue(shape, BATCH_DIM, BATCH_VALUE);
+        case 4:
+          return shape;
+        default:
+          throw new IllegalArgumentException(
+              getShapeInfoMessage() + "The provided image shape is " + Arrays.toString(shape));
+      }
+    }
+
+    @Override
+    String getShapeInfoMessage() {
+      return "The shape of a RGB image should be (h, w, c) or (1, h, w, c), and channels"
+          + " representing R, G, B in order. ";
+    }
+
+    @Override
+    Config toBitmapConfig() {
+      return Config.ARGB_8888;
+    }
+  },
+
+  /** Each pixel is a single element representing only the amount of light. */
+  GRAYSCALE {
+
+    // The channel axis should always be 1 for grayscale images.
+    private static final int CHANNEL_VALUE = 1;
+
+    @Override
+    Bitmap convertTensorBufferToBitmap(TensorBuffer buffer) {
+      return ImageConversions.convertGrayscaleTensorBufferToBitmap(buffer);
+    }
+
+    @Override
+    int getChannelValue() {
+      return CHANNEL_VALUE;
+    }
+
+    @Override
+    int[] getNormalizedShape(int[] shape) {
+      switch (shape.length) {
+          // The shape is in (h, w) format.
+        case 2:
+          int[] shapeWithBatch = insertValue(shape, BATCH_DIM, BATCH_VALUE);
+          return insertValue(shapeWithBatch, CHANNEL_DIM, CHANNEL_VALUE);
+        case 4:
+          return shape;
+        default:
+          // (1, h, w) and (h, w, 1) are potential grayscale image shapes. However, since they
+          // both have three dimensions, it will require extra info to differentiate between them.
+          // Since we haven't encountered real use cases of these two shapes, they are not supported
+          // at this moment to avoid confusion. We may want to revisit it in the future.
+          throw new IllegalArgumentException(
+              getShapeInfoMessage() + "The provided image shape is " + Arrays.toString(shape));
+      }
+    }
+
+    @Override
+    String getShapeInfoMessage() {
+      return "The shape of a grayscale image should be (h, w) or (1, h, w, 1). ";
+    }
+
+    @Override
+    Config toBitmapConfig() {
+      return Config.ALPHA_8;
+    }
+  };
+
+  private static final int BATCH_DIM = 0; // The first element of the normalizaed shape.
+  private static final int BATCH_VALUE = 1; // The batch axis should always be one.
+  private static final int HEIGHT_DIM = 1; // The second element of the normalizaed shape.
+  private static final int WIDTH_DIM = 2; // The third element of the normalizaed shape.
+  private static final int CHANNEL_DIM = 3; // The fourth element of the normalizaed shape.
+
+  /**
+   * Converts a bitmap configuration into the corresponding color space type.
+   *
+   * @throws IllegalArgumentException if the config is unsupported
+   */
+  static ColorSpaceType fromBitmapConfig(Config config) {
+    switch (config) {
+      case ARGB_8888:
+        return ColorSpaceType.RGB;
+      case ALPHA_8:
+        return ColorSpaceType.GRAYSCALE;
+      default:
+        throw new IllegalArgumentException(
+            "Bitmap configuration: " + config + ", is not supported yet.");
+    }
+  }
+
+  /**
+   * Verifies if the given shape matches the color space type.
+   *
+   * @throws IllegalArgumentException if {@code shape} does not match the color space type
+   */
+  void assertShape(int[] shape) {
+    int[] normalizedShape = getNormalizedShape(shape);
+    checkArgument(
+        isValidNormalizedShape(normalizedShape),
+        getShapeInfoMessage() + "The provided image shape is " + Arrays.toString(shape));
+  }
+
+  /**
+   * Converts a {@link TensorBuffer} that represents an image to a Bitmap with the color space type.
+   *
+   * @throws IllegalArgumentException if the shape of buffer does not match the color space type
+   */
+  abstract Bitmap convertTensorBufferToBitmap(TensorBuffer buffer);
+
+  /**
+   * Returns the width of the given shape corresponding to the color space type.
+   *
+   * @throws IllegalArgumentException if {@code shape} does not match the color space type
+   */
+  int getWidth(int[] shape) {
+    assertShape(shape);
+    return getNormalizedShape(shape)[WIDTH_DIM];
+  }
+
+  /**
+   * Returns the height of the given shape corresponding to the color space type.
+   *
+   * @throws IllegalArgumentException if {@code shape} does not match the color space type
+   */
+  int getHeight(int[] shape) {
+    assertShape(shape);
+    return getNormalizedShape(shape)[HEIGHT_DIM];
+  }
+
+  abstract int getChannelValue();
+
+  /**
+   * Gets the normalized shape in the form of (1, h, w, c). Sometimes, a given shape may not have
+   * batch or channel axis.
+   */
+  abstract int[] getNormalizedShape(int[] shape);
+
+  abstract String getShapeInfoMessage();
+
+  /** Converts the color space type to the corresponding bitmap config. */
+  abstract Config toBitmapConfig();
+
+  /** Inserts a value at the specified position and return the new array. */
+  private static int[] insertValue(int[] array, int pos, int value) {
+    int[] newArray = new int[array.length + 1];
+    for (int i = 0; i < pos; i++) {
+      newArray[i] = array[i];
+    }
+    newArray[pos] = value;
+    for (int i = pos + 1; i < newArray.length; i++) {
+      newArray[i] = array[i - 1];
+    }
+    return newArray;
+  }
+
+  protected boolean isValidNormalizedShape(int[] shape) {
+    if (shape[BATCH_DIM] == BATCH_VALUE
+        && shape[HEIGHT_DIM] > 0
+        && shape[WIDTH_DIM] > 0
+        && shape[CHANNEL_DIM] == getChannelValue()) {
+      return true;
+    }
+    return false;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageContainer.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageContainer.java
new file mode 100644
index 0000000..1a145de
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageContainer.java
@@ -0,0 +1,55 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import android.graphics.Bitmap;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Handles image conversion across different image types.
+ *
+ * <p>An {@link ImageContainer} should support the conversion between the underlying image format to
+ * the following image types:
+ *
+ * <ul>
+ *   <li>{@link Bitmap}
+ *   <li>{@link TensorBuffer} of the specified data type.
+ * </ul>
+ */
+interface ImageContainer {
+
+  /** Performs deep copy of the {@link ImageContainer}. */
+  ImageContainer clone();
+
+  /** Returns the width of the image. */
+  int getWidth();
+
+  /** Returns the height of the image. */
+  int getHeight();
+
+  /** Gets the {@link Bitmap} representation of the underlying image format. */
+  Bitmap getBitmap();
+
+  /**
+   * Gets the {@link TensorBuffer} representation with the specific {@code dataType} of the
+   * underlying image format.
+   */
+  TensorBuffer getTensorBuffer(DataType dataType);
+
+  /** Returns the color space type of the image. */
+  ColorSpaceType getColorSpaceType();
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageConversions.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageConversions.java
new file mode 100644
index 0000000..d6e567a
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageConversions.java
@@ -0,0 +1,146 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import android.graphics.Bitmap;
+import android.graphics.Color;
+import java.nio.ByteBuffer;
+import java.nio.ByteOrder;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Implements some stateless image conversion methods.
+ *
+ * <p>This class is an internal helper for {@link org.tensorflow.lite.support.image}.
+ */
+class ImageConversions {
+
+  /**
+   * Converts a {@link TensorBuffer} that represents a RGB image to an ARGB_8888 Bitmap.
+   *
+   * <p>Data in buffer will be converted into integer to match the Bitmap API.
+   *
+   * @param buffer a RGB image. Its shape should be either (h, w, 3) or (1, h, w, 3)
+   * @throws IllegalArgumentException if the shape of buffer is neither (h, w, 3) nor (1, h, w, 3)
+   */
+  static Bitmap convertRgbTensorBufferToBitmap(TensorBuffer buffer) {
+    int[] shape = buffer.getShape();
+    ColorSpaceType rgb = ColorSpaceType.RGB;
+    rgb.assertShape(shape);
+
+    int h = rgb.getHeight(shape);
+    int w = rgb.getWidth(shape);
+    Bitmap bitmap = Bitmap.createBitmap(w, h, rgb.toBitmapConfig());
+
+    // TODO(b/138904567): Find a way to avoid creating multiple intermediate buffers every time.
+    int[] intValues = new int[w * h];
+    int[] rgbValues = buffer.getIntArray();
+    for (int i = 0, j = 0; i < intValues.length; i++) {
+      int r = rgbValues[j++];
+      int g = rgbValues[j++];
+      int b = rgbValues[j++];
+      intValues[i] = Color.rgb(r, g, b);
+    }
+    bitmap.setPixels(intValues, 0, w, 0, 0, w, h);
+
+    return bitmap;
+  }
+
+  /**
+   * Converts a {@link TensorBuffer} that represents a grayscale image to an ALPHA_8 Bitmap.
+   *
+   * <p>Data in buffer will be converted into integer to match the Bitmap API.
+   *
+   * @param buffer a grayscale image. Its shape should be either (h, w) or (1, h, w)
+   * @throws IllegalArgumentException if the shape of buffer is neither (h, w) nor (1, h, w, 1)
+   */
+  static Bitmap convertGrayscaleTensorBufferToBitmap(TensorBuffer buffer) {
+    // Convert buffer into Uint8 as needed.
+    TensorBuffer uint8Buffer =
+        buffer.getDataType() == DataType.UINT8
+            ? buffer
+            : TensorBuffer.createFrom(buffer, DataType.UINT8);
+
+    int[] shape = uint8Buffer.getShape();
+    ColorSpaceType grayscale = ColorSpaceType.GRAYSCALE;
+    grayscale.assertShape(shape);
+
+    // Even though `Bitmap.createBitmap(int[] colors, int width, int height, Bitmap.Config config)`
+    // seems to work for internal Android testing framework, but it actually doesn't work for the
+    // real Android environment.
+    //
+    // The only reliable way to create an ALPHA_8 Bitmap is to use `copyPixelsFromBuffer()` to load
+    // the pixels from a ByteBuffer, and then use `copyPixelsToBuffer` to read out.
+    // Note: for ALPHA_8 Bitmap, methods such as, `setPixels()` and `getPixels()` do not work.
+    Bitmap bitmap =
+        Bitmap.createBitmap(
+            grayscale.getWidth(shape), grayscale.getHeight(shape), grayscale.toBitmapConfig());
+    uint8Buffer.getBuffer().rewind();
+    bitmap.copyPixelsFromBuffer(uint8Buffer.getBuffer());
+    return bitmap;
+  }
+
+  /**
+   * Converts an Image in a Bitmap to a TensorBuffer (3D Tensor: Width-Height-Channel) whose memory
+   * is already allocated, or could be dynamically allocated.
+   *
+   * @param bitmap The Bitmap object representing the image. Currently we only support ARGB_8888
+   *     config.
+   * @param buffer The destination of the conversion. Needs to be created in advance. If it's
+   *     fixed-size, its flat size should be w*h*3.
+   * @throws IllegalArgumentException if the buffer is fixed-size, but the size doesn't match.
+   */
+  static void convertBitmapToTensorBuffer(Bitmap bitmap, TensorBuffer buffer) {
+    int w = bitmap.getWidth();
+    int h = bitmap.getHeight();
+    int[] intValues = new int[w * h];
+    bitmap.getPixels(intValues, 0, w, 0, 0, w, h);
+    // TODO(b/138904567): Find a way to avoid creating multiple intermediate buffers every time.
+    int flatSize = w * h * 3;
+    int[] shape = new int[] {h, w, 3};
+    switch (buffer.getDataType()) {
+      case UINT8:
+        byte[] byteArr = new byte[w * h * 3];
+        for (int i = 0, j = 0; i < intValues.length; i++) {
+          byteArr[j++] = (byte) ((intValues[i] >> 16) & 0xff);
+          byteArr[j++] = (byte) ((intValues[i] >> 8) & 0xff);
+          byteArr[j++] = (byte) (intValues[i] & 0xff);
+        }
+        ByteBuffer byteBuffer = ByteBuffer.allocateDirect(flatSize);
+        byteBuffer.order(ByteOrder.nativeOrder());
+        byteBuffer.put(byteArr);
+        buffer.loadBuffer(byteBuffer, shape);
+        break;
+      case FLOAT32:
+        float[] floatArr = new float[w * h * 3];
+        for (int i = 0, j = 0; i < intValues.length; i++) {
+          floatArr[j++] = (float) ((intValues[i] >> 16) & 0xff);
+          floatArr[j++] = (float) ((intValues[i] >> 8) & 0xff);
+          floatArr[j++] = (float) (intValues[i] & 0xff);
+        }
+        buffer.loadArray(floatArr, shape);
+        break;
+      default:
+        // Should never happen.
+        throw new IllegalStateException(
+            "The type of TensorBuffer, " + buffer.getBuffer() + ", is unsupported.");
+    }
+  }
+
+  // Hide the constructor as the class is static.
+  private ImageConversions() {}
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageOperator.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageOperator.java
new file mode 100644
index 0000000..1e54663
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageOperator.java
@@ -0,0 +1,43 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import android.graphics.PointF;
+import org.tensorflow.lite.support.common.Operator;
+
+/** Operates a TensorImage object. Used in ImageProcessor. */
+public interface ImageOperator extends Operator<TensorImage> {
+  /** @see org.tensorflow.lite.support.common.Operator#apply(java.lang.Object) */
+  @Override
+  TensorImage apply(TensorImage image);
+
+  /** Computes the width of the expected output image when input image size is given. */
+  int getOutputImageWidth(int inputImageHeight, int inputImageWidth);
+
+  /** Computes the height of the expected output image when input image size is given. */
+  int getOutputImageHeight(int inputImageHeight, int inputImageWidth);
+
+  /**
+   * Transforms a point from coordinates system of the result image back to the one of the input
+   * image.
+   *
+   * @param point the point from the result coordinates system.
+   * @param inputImageHeight the height of input image.
+   * @param inputImageWidth the width of input image.
+   * @return the point with the coordinates from the coordinates system of the input image.
+   */
+  PointF inverseTransform(PointF point, int inputImageHeight, int inputImageWidth);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageProcessor.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageProcessor.java
new file mode 100644
index 0000000..e1ef130
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ImageProcessor.java
@@ -0,0 +1,198 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import android.graphics.PointF;
+import android.graphics.RectF;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.ListIterator;
+import org.tensorflow.lite.support.common.Operator;
+import org.tensorflow.lite.support.common.SequentialProcessor;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+import org.tensorflow.lite.support.common.TensorOperator;
+import org.tensorflow.lite.support.image.ops.Rot90Op;
+import org.tensorflow.lite.support.image.ops.TensorOperatorWrapper;
+
+/**
+ * ImageProcessor is a helper class for preprocessing and postprocessing {@link TensorImage}. It
+ * could transform a {@link TensorImage} to another by executing a chain of {@link ImageOperator}.
+ *
+ * <p>Example Usage:
+ *
+ * <pre>
+ *   ImageProcessor processor = new ImageProcessor.Builder()
+ *       .add(new ResizeOp(224, 224, ResizeMethod.NEAREST_NEIGHBOR)
+ *       .add(new Rot90Op())
+ *       .add(new NormalizeOp(127.5f, 127.5f))
+ *       .build();
+ *   TensorImage anotherTensorImage = processor.process(tensorImage);
+ * </pre>
+ *
+ * <p><b>WARNING:</b> Instances of an {@code ImageProcessor} are <b>not</b> thread-safe with {@link
+ * #updateNumberOfRotations}. Updating the number of rotations and then processing images (using
+ * {@link #process}) must be protected from concurrent access. It is recommended to create separate
+ * {@code ImageProcessor} instances for each thread. If multiple threads access a {@code
+ * ImageProcessor} concurrently, it must be synchronized externally.
+ *
+ * @see ImageProcessor.Builder to build a {@link ImageProcessor} instance
+ * @see ImageProcessor#process(TensorImage) to apply the processor on a {@link TensorImage}
+ */
+public class ImageProcessor extends SequentialProcessor<TensorImage> {
+  private ImageProcessor(Builder builder) {
+    super(builder);
+  }
+
+  /**
+   * Transforms a point from coordinates system of the result image back to the one of the input
+   * image.
+   *
+   * @param point the point from the result coordinates system.
+   * @param inputImageHeight the height of input image.
+   * @param inputImageWidth the width of input image.
+   * @return the point with the coordinates from the coordinates system of the input image.
+   */
+  public PointF inverseTransform(PointF point, int inputImageHeight, int inputImageWidth) {
+    List<Integer> widths = new ArrayList<>();
+    List<Integer> heights = new ArrayList<>();
+    int currentWidth = inputImageWidth;
+    int currentHeight = inputImageHeight;
+    for (Operator<TensorImage> op : operatorList) {
+      widths.add(currentWidth);
+      heights.add(currentHeight);
+      ImageOperator imageOperator = (ImageOperator) op;
+      int newHeight = imageOperator.getOutputImageHeight(currentHeight, currentWidth);
+      int newWidth = imageOperator.getOutputImageWidth(currentHeight, currentWidth);
+      currentHeight = newHeight;
+      currentWidth = newWidth;
+    }
+    ListIterator<Operator<TensorImage>> opIterator = operatorList.listIterator(operatorList.size());
+    ListIterator<Integer> widthIterator = widths.listIterator(widths.size());
+    ListIterator<Integer> heightIterator = heights.listIterator(heights.size());
+    while (opIterator.hasPrevious()) {
+      ImageOperator imageOperator = (ImageOperator) opIterator.previous();
+      int height = heightIterator.previous();
+      int width = widthIterator.previous();
+      point = imageOperator.inverseTransform(point, height, width);
+    }
+    return point;
+  }
+
+  /**
+   * Transforms a rectangle from coordinates system of the result image back to the one of the input
+   * image.
+   *
+   * @param rect the rectangle from the result coordinates system.
+   * @param inputImageHeight the height of input image.
+   * @param inputImageWidth the width of input image.
+   * @return the rectangle with the coordinates from the coordinates system of the input image.
+   */
+  public RectF inverseTransform(RectF rect, int inputImageHeight, int inputImageWidth) {
+    // when rotation is involved, corner order may change - top left changes to bottom right, .etc
+    PointF p1 =
+        inverseTransform(new PointF(rect.left, rect.top), inputImageHeight, inputImageWidth);
+    PointF p2 =
+        inverseTransform(new PointF(rect.right, rect.bottom), inputImageHeight, inputImageWidth);
+    return new RectF(
+        Math.min(p1.x, p2.x), Math.min(p1.y, p2.y), Math.max(p1.x, p2.x), Math.max(p1.y, p2.y));
+  }
+
+  /**
+   * The Builder to create an ImageProcessor, which could be executed later.
+   *
+   * @see #add(TensorOperator) to add a general TensorOperator
+   * @see #add(ImageOperator) to add an ImageOperator
+   * @see #build() complete the building process and get a built Processor
+   */
+  public static class Builder extends SequentialProcessor.Builder<TensorImage> {
+    public Builder() {
+      super();
+    }
+
+    /**
+     * Adds an {@link ImageOperator} into the Operator chain.
+     *
+     * @param op the Operator instance to be executed then
+     */
+    public Builder add(ImageOperator op) {
+      super.add(op);
+      return this;
+    }
+
+    /**
+     * Adds a {@link TensorOperator} into the Operator chain. In execution, the processor calls
+     * {@link TensorImage#getTensorBuffer()} to transform the {@link TensorImage} by transforming
+     * the underlying {@link org.tensorflow.lite.support.tensorbuffer.TensorBuffer}.
+     *
+     * @param op the Operator instance to be executed then
+     */
+    public Builder add(TensorOperator op) {
+      return add(new TensorOperatorWrapper(op));
+    }
+
+    /** Completes the building process and gets the {@link ImageProcessor} instance. */
+    @Override
+    public ImageProcessor build() {
+      return new ImageProcessor(this);
+    }
+  }
+
+  /**
+   * Updates the number of rotations for the first {@link Rot90Op} in this {@link ImageProcessor}.
+   *
+   * <p><b>WARNING:</b>this method is <b>not</b> thread-safe. Updating the number of rotations and
+   * then processing images (using {@link #process}) must be protected from concurrent access with
+   * additional synchronization.
+   *
+   * @param k the number of rotations
+   * @throws IllegalStateException if {@link Rot90Op} has not been added to this {@link
+   *     ImageProcessor}
+   */
+  public void updateNumberOfRotations(int k) {
+    updateNumberOfRotations(k, /*occurrence=*/ 0);
+  }
+
+  /**
+   * Updates the number of rotations for the {@link Rot90Op} specified by {@code occurrence} in this
+   * {@link ImageProcessor}.
+   *
+   * <p><b>WARNING:</b>this method is <b>not</b> thread-safe. Updating the number of rotations and
+   * then processing images (using {@link #process}) must be protected from concurrent access with
+   * additional synchronization.
+   *
+   * @param k the number of rotations
+   * @param occurrence the index of perticular {@link Rot90Op} in this {@link ImageProcessor}. For
+   *     example, if the second {@link Rot90Op} needs to be updated, {@code occurrence} should be
+   *     set to 1.
+   * @throws IndexOutOfBoundsException if {@code occurrence} is negative or is not less than the
+   *     number of {@link Rot90Op} in this {@link ImageProcessor}
+   * @throws IllegalStateException if {@link Rot90Op} has not been added to this {@link
+   *     ImageProcessor}
+   */
+  public synchronized void updateNumberOfRotations(int k, int occurrence) {
+    SupportPreconditions.checkState(
+        operatorIndex.containsKey(Rot90Op.class.getName()),
+        "The Rot90Op has not been added to the ImageProcessor.");
+
+    List<Integer> indexes = operatorIndex.get(Rot90Op.class.getName());
+    SupportPreconditions.checkElementIndex(occurrence, indexes.size(), "occurrence");
+
+    // The index of the Rot90Op to be replaced in operatorList.
+    int index = indexes.get(occurrence);
+    Rot90Op newRot = new Rot90Op(k);
+    operatorList.set(index, newRot);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/TensorBufferContainer.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/TensorBufferContainer.java
new file mode 100644
index 0000000..d047a8e
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/TensorBufferContainer.java
@@ -0,0 +1,90 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import android.graphics.Bitmap;
+import android.util.Log;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/** Holds a {@link TensorBuffer} and converts it to other image formats as needed. */
+final class TensorBufferContainer implements ImageContainer {
+
+  private final TensorBuffer buffer;
+  private final ColorSpaceType colorSpaceType;
+  private static final String TAG = TensorBufferContainer.class.getSimpleName();
+
+  /**
+   * Creates a {@link TensorBufferContainer} object with the specified {@link
+   * TensorImage#ColorSpaceType}.
+   *
+   * @throws IllegalArgumentException if the shape of the {@link TensorBuffer} does not match the
+   *     specified color space type
+   */
+  static TensorBufferContainer create(TensorBuffer buffer, ColorSpaceType colorSpaceType) {
+    return new TensorBufferContainer(buffer, colorSpaceType);
+  }
+
+  private TensorBufferContainer(TensorBuffer buffer, ColorSpaceType colorSpaceType) {
+    colorSpaceType.assertShape(buffer.getShape());
+    this.buffer = buffer;
+    this.colorSpaceType = colorSpaceType;
+  }
+
+  @Override
+  public TensorBufferContainer clone() {
+    return create(TensorBuffer.createFrom(buffer, buffer.getDataType()), colorSpaceType);
+  }
+
+  @Override
+  public Bitmap getBitmap() {
+    if (buffer.getDataType() != DataType.UINT8) {
+      // Print warning instead of throwing an exception. When using float models, users may want to
+      // convert the resulting float image into Bitmap. That's fine to do so, as long as they are
+      // aware of the potential accuracy lost when casting to uint8.
+      Log.w(
+          TAG,
+          "<Warning> TensorBufferContainer is holding a non-uint8 image. The conversion to Bitmap"
+              + " will cause numeric casting and clamping on the data value.");
+    }
+
+    return colorSpaceType.convertTensorBufferToBitmap(buffer);
+  }
+
+  @Override
+  public TensorBuffer getTensorBuffer(DataType dataType) {
+    // If the data type of buffer is desired, return it directly. Not making a defensive copy for
+    // performance considerations. During image processing, users may need to set and get the
+    // TensorBuffer many times.
+    // Otherwise, create another one with the expected data type.
+    return buffer.getDataType() == dataType ? buffer : TensorBuffer.createFrom(buffer, dataType);
+  }
+
+  @Override
+  public int getWidth() {
+    return colorSpaceType.getWidth(buffer.getShape());
+  }
+
+  @Override
+  public int getHeight() {
+    return colorSpaceType.getHeight(buffer.getShape());
+  }
+
+  @Override
+  public ColorSpaceType getColorSpaceType() {
+    return colorSpaceType;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/TensorImage.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/TensorImage.java
new file mode 100644
index 0000000..96cae71
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/TensorImage.java
@@ -0,0 +1,312 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image;
+
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkArgument;
+
+import android.graphics.Bitmap;
+import java.nio.ByteBuffer;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * TensorImage is the wrapper class for Image object. When using image processing utils in
+ * TFLite.support library, it's common to convert image objects in variant types to TensorImage at
+ * first.
+ *
+ * <p>At present, only RGB images are supported, and the A channel is always ignored.
+ *
+ * <p>Details of data storage: a {@link TensorImage} object may have 2 potential sources of truth: a
+ * {@link Bitmap} or a {@link TensorBuffer}. {@link TensorImage} maintains the state and only
+ * converts one to the other when needed. A typical use case of {@link TensorImage} is to first load
+ * a {@link Bitmap} image, then process it using {@link ImageProcessor}, and finally get the
+ * underlying {@link ByteBuffer} of the {@link TensorBuffer} and feed it into the TFLite
+ * interpreter.
+ *
+ * <p>IMPORTANT: to achieve the best performance, {@link TensorImage} avoids copying data whenever
+ * it's possible. Therefore, it doesn't own its data. Callers should not modify data objects those
+ * are passed to {@link TensorImage#load(Bitmap)} or {@link TensorImage#load(TensorBuffer)}.
+ *
+ * <p>IMPORTANT: all methods are not proved thread-safe.
+ *
+ * @see ImageProcessor which is often used for transforming a {@link TensorImage}.
+ */
+// TODO(b/138907116): Support loading images from TensorBuffer with properties.
+// TODO(b/138905544): Support directly loading RGBBytes, YUVBytes and other types if necessary.
+public class TensorImage {
+
+  private final DataType dataType;
+  private ImageContainer container = null;
+
+  /**
+   * Initializes a {@link TensorImage} object.
+   *
+   * <p>Note: the data type of this {@link TensorImage} is {@link DataType#UINT8}. Use {@link
+   * #TensorImage(DataType)} if other data types are preferred.
+   */
+  public TensorImage() {
+    this(DataType.UINT8);
+  }
+
+  /**
+   * Initializes a {@link TensorImage} object with the specified data type.
+   *
+   * <p>When getting a {@link TensorBuffer} or a {@link ByteBuffer} from this {@link TensorImage},
+   * such as using {@link #getTensorBuffer} and {@link #getBuffer}, the data values will be
+   * converted to the specified data type.
+   *
+   * <p>Note: the shape of a {@link TensorImage} is not fixed. It can be adjusted to the shape of
+   * the image being loaded to this {@link TensorImage}.
+   *
+   * @param dataType the expected data type of the resulting {@link TensorBuffer}. The type is
+   *     always fixed during the lifetime of the {@link TensorImage}. To convert the data type, use
+   *     {@link #createFrom(TensorImage, DataType)} to create a copy and convert data type at the
+   *     same time.
+   * @throws IllegalArgumentException if {@code dataType} is neither {@link DataType#UINT8} nor
+   *     {@link DataType#FLOAT32}
+   */
+  public TensorImage(DataType dataType) {
+    checkArgument(
+        dataType == DataType.UINT8 || dataType == DataType.FLOAT32,
+        "Illegal data type for TensorImage: Only FLOAT32 and UINT8 are accepted");
+    this.dataType = dataType;
+  }
+
+  /**
+   * Initializes a {@link TensorImage} object of {@link DataType#UINT8} with a {@link Bitmap} .
+   *
+   * @see TensorImage#load(Bitmap) for reusing the object when it's expensive to create objects
+   *     frequently, because every call of {@code fromBitmap} creates a new {@link TensorImage}.
+   */
+  public static TensorImage fromBitmap(Bitmap bitmap) {
+    TensorImage image = new TensorImage();
+    image.load(bitmap);
+    return image;
+  }
+
+  /**
+   * Creates a deep-copy of a given {@link TensorImage} with the desired data type.
+   *
+   * @param src the {@link TensorImage} to copy from
+   * @param dataType the expected data type of newly created {@link TensorImage}
+   * @return a {@link TensorImage} whose data is copied from {@code src} and data type is {@code
+   *     dataType}
+   */
+  public static TensorImage createFrom(TensorImage src, DataType dataType) {
+    TensorImage dst = new TensorImage(dataType);
+    dst.container = src.container.clone();
+    return dst;
+  }
+
+  /**
+   * Loads a {@link Bitmap} image object into this {@link TensorImage}.
+   *
+   * <p>Note: if the {@link TensorImage} has data type other than {@link DataType#UINT8}, numeric
+   * casting and clamping will be applied when calling {@link #getTensorBuffer} and {@link
+   * #getBuffer}, where the {@link Bitmap} will be converted into a {@link TensorBuffer}.
+   *
+   * <p>Important: when loading a bitmap, DO NOT MODIFY the bitmap from the caller side anymore. The
+   * {@link TensorImage} object will rely on the bitmap. It will probably modify the bitmap as well.
+   * In this method, we perform a zero-copy approach for that bitmap, by simply holding its
+   * reference. Use {@code bitmap.copy(bitmap.getConfig(), true)} to create a copy if necessary.
+   *
+   * <p>Note: to get the best performance, please load images in the same shape to avoid memory
+   * re-allocation.
+   *
+   * @throws IllegalArgumentException if {@code bitmap} is not in ARGB_8888
+   */
+  public void load(Bitmap bitmap) {
+    container = BitmapContainer.create(bitmap);
+  }
+
+  /**
+   * Loads a float array as RGB pixels into this {@link TensorImage}, representing the pixels
+   * inside.
+   *
+   * <p>Note: if the {@link TensorImage} has a data type other than {@link DataType#FLOAT32},
+   * numeric casting and clamping will be applied when calling {@link #getTensorBuffer} and {@link
+   * #getBuffer}.
+   *
+   * @param pixels the RGB pixels representing the image
+   * @param shape the shape of the image, should either in form (h, w, 3), or in form (1, h, w, 3)
+   * @throws IllegalArgumentException if the shape is neither (h, w, 3) nor (1, h, w, 3)
+   */
+  public void load(float[] pixels, int[] shape) {
+    TensorBuffer buffer = TensorBuffer.createDynamic(getDataType());
+    buffer.loadArray(pixels, shape);
+    load(buffer);
+  }
+
+  /**
+   * Loads an int array as RGB pixels into this {@link TensorImage}, representing the pixels inside.
+   *
+   * <p>Note: numeric casting and clamping will be applied to convert the values into the data type
+   * of this {@link TensorImage} when calling {@link #getTensorBuffer} and {@link #getBuffer}.
+   *
+   * @param pixels the RGB pixels representing the image
+   * @param shape the shape of the image, should either in form (h, w, 3), or in form (1, h, w, 3)
+   * @throws IllegalArgumentException if the shape is neither (h, w, 3) nor (1, h, w, 3)
+   */
+  public void load(int[] pixels, int[] shape) {
+    TensorBuffer buffer = TensorBuffer.createDynamic(getDataType());
+    buffer.loadArray(pixels, shape);
+    load(buffer);
+  }
+
+  /**
+   * Loads a {@link TensorBuffer} containing pixel values. The color layout should be RGB.
+   *
+   * <p>Note: if the data type of {@code buffer} does not match that of this {@link TensorImage},
+   * numeric casting and clamping will be applied when calling {@link #getTensorBuffer} and {@link
+   * #getBuffer}.
+   *
+   * @param buffer the {@link TensorBuffer} to be loaded. Its shape should be either (h, w, 3) or
+   *     (1, h, w, 3)
+   * @throws IllegalArgumentException if the shape is neither (h, w, 3) nor (1, h, w, 3)
+   */
+  public void load(TensorBuffer buffer) {
+    load(buffer, ColorSpaceType.RGB);
+  }
+
+  /**
+   * Loads a {@link TensorBuffer} containing pixel values with the specific {@link ColorSapceType}.
+   *
+   * <p>Note: if the data type of {@code buffer} does not match that of this {@link TensorImage},
+   * numeric casting and clamping will be applied when calling {@link #getTensorBuffer} and {@link
+   * #getBuffer}.
+   *
+   * @throws IllegalArgumentException if the shape of buffer does not match the color space type
+   * @see ColorSpaceType#assertShape
+   */
+  public void load(TensorBuffer buffer, ColorSpaceType colorSpaceType) {
+    container = TensorBufferContainer.create(buffer, colorSpaceType);
+  }
+
+  /**
+   * Returns a {@link Bitmap} representation of this {@link TensorImage}.
+   *
+   * <p>Numeric casting and clamping will be applied if the stored data is not uint8.
+   *
+   * <p>Note that, the reliable way to get pixels from an {@code ALPHA_8} Bitmap is to use {@code
+   * copyPixelsToBuffer}. Bitmap methods such as, `setPixels()` and `getPixels` do not work.
+   *
+   * <p>Important: it's only a reference. DO NOT MODIFY. We don't create a copy here for performance
+   * concern, but if modification is necessary, please make a copy.
+   *
+   * @return a reference to a {@link Bitmap} in {@code ARGB_8888} config ("A" channel is always
+   *     opaque) or in {@code ALPHA_8}, depending on the {@link ColorSpaceType} of this {@link
+   *     TensorBuffer}.
+   * @throws IllegalStateException if the {@link TensorImage} never loads data
+   */
+  public Bitmap getBitmap() {
+    if (container == null) {
+      throw new IllegalStateException("No image has been loaded yet.");
+    }
+
+    return container.getBitmap();
+  }
+
+  /**
+   * Returns a {@link ByteBuffer} representation of this {@link TensorImage} with the expected data
+   * type.
+   *
+   * <p>Numeric casting and clamping will be applied if the stored data is different from the data
+   * type of the {@link TensorImage}.
+   *
+   * <p>Important: it's only a reference. DO NOT MODIFY. We don't create a copy here for performance
+   * concern, but if modification is necessary, please make a copy.
+   *
+   * <p>It's essentially a short cut for {@code getTensorBuffer().getBuffer()}.
+   *
+   * @return a reference to a {@link ByteBuffer} which holds the image data
+   * @throws IllegalStateException if the {@link TensorImage} never loads data
+   */
+  public ByteBuffer getBuffer() {
+    return getTensorBuffer().getBuffer();
+  }
+
+  /**
+   * Returns a {@link TensorBuffer} representation of this {@link TensorImage} with the expected
+   * data type.
+   *
+   * <p>Numeric casting and clamping will be applied if the stored data is different from the data
+   * type of the {@link TensorImage}.
+   *
+   * <p>Important: it's only a reference. DO NOT MODIFY. We don't create a copy here for performance
+   * concern, but if modification is necessary, please make a copy.
+   *
+   * @return a reference to a {@link TensorBuffer} which holds the image data
+   * @throws IllegalStateException if the {@link TensorImage} never loads data
+   */
+  public TensorBuffer getTensorBuffer() {
+    if (container == null) {
+      throw new IllegalStateException("No image has been loaded yet.");
+    }
+
+    return container.getTensorBuffer(dataType);
+  }
+
+  /**
+   * Gets the data type of this {@link TensorImage}.
+   *
+   * @return a data type. Currently only {@link DataType#UINT8} and {@link DataType#FLOAT32} are
+   *     supported.
+   */
+  public DataType getDataType() {
+    return dataType;
+  }
+
+  /**
+   * Gets the color space type of this {@link TensorImage}.
+   *
+   * @throws IllegalStateException if the {@link TensorImage} never loads data
+   */
+  public ColorSpaceType getColorSpaceType() {
+    if (container == null) {
+      throw new IllegalStateException("No image has been loaded yet.");
+    }
+
+    return container.getColorSpaceType();
+  }
+
+  /**
+   * Gets the image width.
+   *
+   * @throws IllegalStateException if the {@link TensorImage} never loads data
+   * @throws IllegalArgumentException if the underlying data is corrupted
+   */
+  public int getWidth() {
+    if (container == null) {
+      throw new IllegalStateException("No image has been loaded yet.");
+    }
+
+    return container.getWidth();
+  }
+
+  /**
+   * Gets the image height.
+   *
+   * @throws IllegalStateException if the {@link TensorImage} never loads data
+   * @throws IllegalArgumentException if the underlying data is corrupted
+   */
+  public int getHeight() {
+    if (container == null) {
+      throw new IllegalStateException("No image has been loaded yet.");
+    }
+
+    return container.getHeight();
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/ResizeOp.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/ResizeOp.java
new file mode 100644
index 0000000..35606dd
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/ResizeOp.java
@@ -0,0 +1,89 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image.ops;
+
+import android.graphics.Bitmap;
+import android.graphics.PointF;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.support.image.ImageOperator;
+import org.tensorflow.lite.support.image.TensorImage;
+
+/**
+ * As a computation unit for processing images, it can resize an image to user-specified size.
+ *
+ * <p>It interpolates pixels when image is stretched, and discards pixels when image is compressed.
+ *
+ * @see ResizeWithCropOrPadOp for resizing without content distortion.
+ */
+public class ResizeOp implements ImageOperator {
+
+  /** Algorithms for resizing. */
+  public enum ResizeMethod {
+    BILINEAR,
+    NEAREST_NEIGHBOR
+  }
+
+  private final int targetHeight;
+  private final int targetWidth;
+  private final boolean useBilinear;
+
+  /**
+   * Creates a ResizeOp which can resize images to specified size in specified method.
+   *
+   * @param targetHeight: The expected height of resized image.
+   * @param targetWidth: The expected width of resized image.
+   * @param resizeMethod: The algorithm to use for resizing. Options: {@link ResizeMethod}
+   */
+  public ResizeOp(int targetHeight, int targetWidth, ResizeMethod resizeMethod) {
+    this.targetHeight = targetHeight;
+    this.targetWidth = targetWidth;
+    useBilinear = (resizeMethod == ResizeMethod.BILINEAR);
+  }
+
+  /**
+   * Applies the defined resizing on given image and returns the result.
+   *
+   * <p>Note: the content of input {@code image} will change, and {@code image} is the same instance
+   * with the output.
+   *
+   * @param image input image.
+   * @return output image.
+   */
+  @Override
+  @NonNull
+  public TensorImage apply(@NonNull TensorImage image) {
+    Bitmap scaled =
+        Bitmap.createScaledBitmap(image.getBitmap(), targetWidth, targetHeight, useBilinear);
+    image.load(scaled);
+    return image;
+  }
+
+  @Override
+  public int getOutputImageHeight(int inputImageHeight, int inputImageWidth) {
+    return targetHeight;
+  }
+
+  @Override
+  public int getOutputImageWidth(int inputImageHeight, int inputImageWidth) {
+    return targetWidth;
+  }
+
+  @Override
+  public PointF inverseTransform(PointF point, int inputImageHeight, int inputImageWidth) {
+    return new PointF(
+        point.x * inputImageWidth / targetWidth, point.y * inputImageHeight / targetHeight);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/ResizeWithCropOrPadOp.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/ResizeWithCropOrPadOp.java
new file mode 100644
index 0000000..404429e
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/ResizeWithCropOrPadOp.java
@@ -0,0 +1,125 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image.ops;
+
+import android.graphics.Bitmap;
+import android.graphics.Bitmap.Config;
+import android.graphics.Canvas;
+import android.graphics.PointF;
+import android.graphics.Rect;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.support.image.ImageOperator;
+import org.tensorflow.lite.support.image.TensorImage;
+
+/**
+ * As a computation unit for processing images, it could resize image to predefined size.
+ *
+ * <p>It will not stretch or compress the content of image. However, to fit the new size, it crops
+ * or pads pixels. When it crops image, it performs a center-crop; when it pads pixels, it performs
+ * a zero-padding.
+ *
+ * @see ResizeOp for reszing images while stretching / compressing the content.
+ */
+public class ResizeWithCropOrPadOp implements ImageOperator {
+  private final int targetHeight;
+  private final int targetWidth;
+  private final Bitmap output;
+
+  /**
+   * Creates a ResizeWithCropOrPadOp which could crop/pad images to specified size. It adopts
+   * center-crop and zero-padding.
+   *
+   * @param targetHeight: The expected height of cropped/padded image.
+   * @param targetWidth: The expected width of cropped/padded image.
+   */
+  public ResizeWithCropOrPadOp(int targetHeight, int targetWidth) {
+    this.targetHeight = targetHeight;
+    this.targetWidth = targetWidth;
+    output = Bitmap.createBitmap(this.targetWidth, this.targetHeight, Config.ARGB_8888);
+  }
+
+  /**
+   * Applies the defined resizing with cropping or/and padding on given image and returns the
+   * result.
+   *
+   * <p>Note: the content of input {@code image} will change, and {@code image} is the same instance
+   * with the output.
+   *
+   * @param image input image.
+   * @return output image.
+   */
+  @Override
+  @NonNull
+  public TensorImage apply(@NonNull TensorImage image) {
+    Bitmap input = image.getBitmap();
+    int srcL;
+    int srcR;
+    int srcT;
+    int srcB;
+    int dstL;
+    int dstR;
+    int dstT;
+    int dstB;
+    int w = input.getWidth();
+    int h = input.getHeight();
+    if (targetWidth > w) { // padding
+      srcL = 0;
+      srcR = w;
+      dstL = (targetWidth - w) / 2;
+      dstR = dstL + w;
+    } else { // cropping
+      dstL = 0;
+      dstR = targetWidth;
+      srcL = (w - targetWidth) / 2;
+      srcR = srcL + targetWidth;
+    }
+    if (targetHeight > h) { // padding
+      srcT = 0;
+      srcB = h;
+      dstT = (targetHeight - h) / 2;
+      dstB = dstT + h;
+    } else { // cropping
+      dstT = 0;
+      dstB = targetHeight;
+      srcT = (h - targetHeight) / 2;
+      srcB = srcT + targetHeight;
+    }
+    Rect src = new Rect(srcL, srcT, srcR, srcB);
+    Rect dst = new Rect(dstL, dstT, dstR, dstB);
+    new Canvas(output).drawBitmap(input, src, dst, null);
+    image.load(output);
+    return image;
+  }
+
+  @Override
+  public int getOutputImageHeight(int inputImageHeight, int inputImageWidth) {
+    return targetHeight;
+  }
+
+  @Override
+  public int getOutputImageWidth(int inputImageHeight, int inputImageWidth) {
+    return targetWidth;
+  }
+
+  @Override
+  public PointF inverseTransform(PointF point, int inputImageHeight, int inputImageWidth) {
+    return transformImpl(point, targetHeight, targetWidth, inputImageHeight, inputImageWidth);
+  }
+
+  private static PointF transformImpl(PointF point, int srcH, int srcW, int dstH, int dstW) {
+    return new PointF(point.x + (dstW - srcW) / 2, point.y + (dstH - srcH) / 2);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/Rot90Op.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/Rot90Op.java
new file mode 100644
index 0000000..2fa2293
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/Rot90Op.java
@@ -0,0 +1,103 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image.ops;
+
+import android.graphics.Bitmap;
+import android.graphics.Matrix;
+import android.graphics.PointF;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.support.image.ImageOperator;
+import org.tensorflow.lite.support.image.TensorImage;
+
+/** Rotates image counter-clockwise. */
+public class Rot90Op implements ImageOperator {
+
+  private final int numRotation;
+
+  /** Creates a Rot90 Op which will rotate image by 90 degree counter-clockwise. */
+  public Rot90Op() {
+    this(1);
+  }
+
+  /**
+   * Creates a Rot90 Op which will rotate image by 90 degree for {@code k} times counter-clockwise.
+   *
+   * @param k: The number of times the image is rotated by 90 degrees. If it's positive, the image
+   *     will be rotated counter-clockwise. If it's negative, the op will rotate image clockwise.
+   */
+  public Rot90Op(int k) {
+    numRotation = k % 4;
+  }
+
+  /**
+   * Applies the defined rotation on given image and returns the result.
+   *
+   * <p>Note: the content of input {@code image} will change, and {@code image} is the same instance
+   * with the output.
+   *
+   * @param image input image.
+   * @return output image.
+   */
+  @NonNull
+  @Override
+  public TensorImage apply(@NonNull TensorImage image) {
+    Bitmap input = image.getBitmap();
+    if (numRotation == 0) {
+      return image;
+    }
+    int w = input.getWidth();
+    int h = input.getHeight();
+    Matrix matrix = new Matrix();
+    matrix.postTranslate(w * 0.5f, h * 0.5f);
+    matrix.postRotate(-90 * numRotation);
+    int newW = (numRotation % 2 == 0) ? w : h;
+    int newH = (numRotation % 2 == 0) ? h : w;
+    matrix.postTranslate(newW * 0.5f, newH * 0.5f);
+    Bitmap output = Bitmap.createBitmap(input, 0, 0, w, h, matrix, false);
+    image.load(output);
+    return image;
+  }
+
+  @Override
+  public int getOutputImageHeight(int inputImageHeight, int inputImageWidth) {
+    return (numRotation % 2 == 0) ? inputImageHeight : inputImageWidth;
+  }
+
+  @Override
+  public int getOutputImageWidth(int inputImageHeight, int inputImageWidth) {
+    return (numRotation % 2 == 0) ? inputImageWidth : inputImageHeight;
+  }
+
+  @Override
+  public PointF inverseTransform(PointF point, int inputImageHeight, int inputImageWidth) {
+    int inverseNumRotation = (4 - numRotation) % 4;
+    int height = getOutputImageHeight(inputImageHeight, inputImageWidth);
+    int width = getOutputImageWidth(inputImageHeight, inputImageWidth);
+    return transformImpl(point, height, width, inverseNumRotation);
+  }
+
+  private static PointF transformImpl(PointF point, int height, int width, int numRotation) {
+    if (numRotation == 0) {
+      return point;
+    } else if (numRotation == 1) {
+      return new PointF(point.y, width - point.x);
+    } else if (numRotation == 2) {
+      return new PointF(width - point.x, height - point.y);
+    } else { // numRotation == 3
+      return new PointF(height - point.y, point.x);
+    }
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/TensorOperatorWrapper.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/TensorOperatorWrapper.java
new file mode 100644
index 0000000..420018d
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/image/ops/TensorOperatorWrapper.java
@@ -0,0 +1,75 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.image.ops;
+
+import android.graphics.PointF;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+import org.tensorflow.lite.support.common.TensorOperator;
+import org.tensorflow.lite.support.image.ImageOperator;
+import org.tensorflow.lite.support.image.TensorImage;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * The adapter that makes a TensorOperator able to run with TensorImage.
+ *
+ * @see org.tensorflow.lite.support.common.TensorOperator
+ * @see org.tensorflow.lite.support.image.TensorImage
+ */
+public class TensorOperatorWrapper implements ImageOperator {
+
+  private final TensorOperator tensorOp;
+
+  /**
+   * Wraps a {@link TensorOperator} object as an {@link ImageOperator}, so that the {@link
+   * TensorOperator} could handle {@link TensorImage} objects by handling its underlying {@link
+   * org.tensorflow.lite.support.tensorbuffer.TensorBuffer}.
+   *
+   * <p>Requirement: The {@code op} should not change coordinate system when applied on an image.
+   *
+   * @param op The created operator.
+   */
+  public TensorOperatorWrapper(TensorOperator op) {
+    tensorOp = op;
+  }
+
+  @Override
+  @NonNull
+  public TensorImage apply(@NonNull TensorImage image) {
+    SupportPreconditions.checkNotNull(image, "Op cannot apply on null image.");
+    TensorBuffer resBuffer = tensorOp.apply(image.getTensorBuffer());
+    // Some ops may change the data type of the underlying TensorBuffer, such as CastOp. Therefore,
+    // need to create a new TensorImage with the correct data type.
+    TensorImage resImage = new TensorImage(resBuffer.getDataType());
+    resImage.load(resBuffer);
+    return resImage;
+  }
+
+  @Override
+  public int getOutputImageHeight(int inputImageHeight, int inputImageWidth) {
+    return inputImageHeight;
+  }
+
+  @Override
+  public int getOutputImageWidth(int inputImageHeight, int inputImageWidth) {
+    return inputImageWidth;
+  }
+
+  @Override
+  public PointF inverseTransform(PointF point, int inputImageHeight, int inputImageWidth) {
+    return point;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/Category.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/Category.java
new file mode 100644
index 0000000..5b043a9
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/Category.java
@@ -0,0 +1,95 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.label;
+
+import java.util.Objects;
+import org.tensorflow.lite.annotations.UsedByReflection;
+
+/**
+ * Category is a util class, contains a label, its display name and a float value as score.
+ * Typically it's used as result of classification tasks.
+ */
+@UsedByReflection("TFLiteSupport/Task")
+public final class Category {
+  private final String label;
+  private final String displayName;
+  private final float score;
+
+  /**
+   * Constructs a {@link Category} object.
+   *
+   * @param displayName the display name of the label, which may be translated for different
+   *     locales. For exmaple, a label, "apple", may be translated into Spanish for display purpose,
+   *     so that the displayName is "manzana".
+   */
+  @UsedByReflection("TFLiteSupport/Task")
+  public static Category create(String label, String displayName, float score) {
+    return new Category(label, displayName, score);
+  }
+
+  @UsedByReflection("TFLiteSupport/Task")
+  /** Constructs a {@link Category} object with an empty displayName. */
+  public Category(String label, float score) {
+    this(label, /*displayName=*/ "", score);
+  }
+
+  private Category(String label, String displayName, float score) {
+    this.label = label;
+    this.displayName = displayName;
+    this.score = score;
+  }
+
+  /** Gets the reference of category's label. */
+  public String getLabel() {
+    return label;
+  }
+
+  /**
+   * Gets the reference of category's displayName, a name in locale of the label.
+   *
+   * <p>The display name can be an empty string if this {@link Category} object is constructed
+   * without displayName, such as when using {@link #Category(String label, float score)}.
+   */
+  public String getDisplayName() {
+    return displayName;
+  }
+
+  /** Gets the score of the category. */
+  public float getScore() {
+    return score;
+  }
+
+  @Override
+  public boolean equals(Object o) {
+    if (o instanceof Category) {
+      Category other = (Category) o;
+      return (other.getLabel().equals(this.label)
+          && other.getDisplayName().equals(this.displayName)
+          && other.getScore() == this.score);
+    }
+    return false;
+  }
+
+  @Override
+  public int hashCode() {
+    return Objects.hash(label, displayName, score);
+  }
+
+  @Override
+  public String toString() {
+    return "<Category \"" + label + "\" (displayName=" + displayName + "\" (score=" + score + ")>";
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/LabelUtil.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/LabelUtil.java
new file mode 100644
index 0000000..840ed5f
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/LabelUtil.java
@@ -0,0 +1,64 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.label;
+
+import android.util.Log;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.List;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/** Label operation utils. */
+public class LabelUtil {
+  /**
+   * Maps an int value tensor to a list of string labels. It takes an array of strings as the
+   * dictionary. Example: if the given tensor is [3, 1, 0], and given labels is ["background",
+   * "apple", "banana", "cherry", "date"], the result will be ["date", "banana", "apple"].
+   *
+   * @param tensorBuffer: A tensor with index values. The values should be non-negative integers,
+   *     and each value {@code x} will be converted to {@code labels[x + offset]}. If the tensor is
+   *     given as a float {@link TensorBuffer}, values will be cast to integers. All values that are
+   *     out of bound will map to empty string.
+   * @param labels: A list of strings, used as a dictionary to look up. The index of the array
+   *     element will be used as the key. To get better performance, use an object that implements
+   *     RandomAccess, such as {@link ArrayList}.
+   * @param offset: The offset value when look up int values in the {@code labels}.
+   * @return the mapped strings. The length of the list is {@link TensorBuffer#getFlatSize}.
+   * @throws IllegalArgumentException if {@code tensorBuffer} or {@code labels} is null.
+   */
+  public static List<String> mapValueToLabels(
+      @NonNull TensorBuffer tensorBuffer, @NonNull List<String> labels, int offset) {
+    SupportPreconditions.checkNotNull(tensorBuffer, "Given tensor should not be null");
+    SupportPreconditions.checkNotNull(labels, "Given labels should not be null");
+    int[] values = tensorBuffer.getIntArray();
+    Log.d("values", Arrays.toString(values));
+    List<String> result = new ArrayList<>();
+    for (int v : values) {
+      int index = v + offset;
+      if (index < 0 || index >= labels.size()) {
+        result.add("");
+      } else {
+        result.add(labels.get(index));
+      }
+    }
+    return result;
+  }
+
+  // Private constructor to prevent initialization.
+  private LabelUtil() {}
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/TensorLabel.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/TensorLabel.java
new file mode 100644
index 0000000..10763a1
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/TensorLabel.java
@@ -0,0 +1,224 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.label;
+
+import android.content.Context;
+import java.nio.ByteBuffer;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * TensorLabel is an util wrapper for TensorBuffers with meaningful labels on an axis.
+ *
+ * <p>For example, an image classification model may have an output tensor with shape as {1, 10},
+ * where 1 is the batch size and 10 is the number of categories. In fact, on the 2nd axis, we could
+ * label each sub-tensor with the name or description of each corresponding category. {@link
+ * TensorLabel} could help converting the plain Tensor in {@link TensorBuffer} into a map from
+ * predefined labels to sub-tensors. In this case, if provided 10 labels for the 2nd axis, {@link
+ * TensorLabel} could convert the original {1, 10} Tensor to a 10 element map, each value of which
+ * is Tensor in shape {} (scalar). Usage example:
+ *
+ * <pre>
+ *   TensorBuffer outputTensor = ...;
+ *   {@literal List<String>} labels = FileUtil.loadLabels(context, labelFilePath);
+ *   // labels the first axis with size greater than one
+ *   TensorLabel labeled = new TensorLabel(labels, outputTensor);
+ *   // If each sub-tensor has effectively size 1, we can directly get a float value
+ *   {@literal Map<String, Float>} probabilities = labeled.getMapWithFloatValue();
+ *   // Or get sub-tensors, when each sub-tensor has elements more than 1
+ *   {@literal Map<String, TensorBuffer>} subTensors = labeled.getMapWithTensorBuffer();
+ * </pre>
+ *
+ * <p>Note: currently we only support tensor-to-map conversion for the first label with size greater
+ * than 1.
+ *
+ * @see org.tensorflow.lite.support.common.FileUtil#loadLabels(Context, String) to load labels from
+ *     a label file (plain text file whose each line is a label) in assets simply.
+ */
+public class TensorLabel {
+  private final Map<Integer, List<String>> axisLabels;
+  private final TensorBuffer tensorBuffer;
+  private final int[] shape;
+
+  /**
+   * Creates a TensorLabel object which is able to label on the axes of multi-dimensional tensors.
+   *
+   * @param axisLabels A map, whose key is axis id (starting from 0) and value is corresponding
+   *     labels. Note: The size of labels should be same with the size of the tensor on that axis.
+   * @param tensorBuffer The TensorBuffer to be labeled.
+   * @throws NullPointerException if {@code axisLabels} or {@code tensorBuffer} is null, or any
+   *     value in {@code axisLabels} is null.
+   * @throws IllegalArgumentException if any key in {@code axisLabels} is out of range (compared to
+   *     the shape of {@code tensorBuffer}, or any value (labels) has different size with the {@code
+   *     tensorBuffer} on the given dimension.
+   */
+  public TensorLabel(
+      @NonNull Map<Integer, List<String>> axisLabels, @NonNull TensorBuffer tensorBuffer) {
+    SupportPreconditions.checkNotNull(axisLabels, "Axis labels cannot be null.");
+    SupportPreconditions.checkNotNull(tensorBuffer, "Tensor Buffer cannot be null.");
+    this.axisLabels = axisLabels;
+    this.tensorBuffer = tensorBuffer;
+    this.shape = tensorBuffer.getShape();
+    for (Map.Entry<Integer, List<String>> entry : axisLabels.entrySet()) {
+      int axis = entry.getKey();
+      SupportPreconditions.checkArgument(
+          axis >= 0 && axis < shape.length, "Invalid axis id: " + axis);
+      SupportPreconditions.checkNotNull(entry.getValue(), "Label list is null on axis " + axis);
+      SupportPreconditions.checkArgument(
+          shape[axis] == entry.getValue().size(),
+          "Label number " + entry.getValue().size() + " mismatch the shape on axis " + axis);
+    }
+  }
+
+  /**
+   * Creates a TensorLabel object which is able to label on one axis of multi-dimensional tensors.
+   *
+   * <p>Note: The labels are applied on the first axis whose size is larger than 1. For example, if
+   * the shape of the tensor is [1, 10, 3], the labels will be applied on axis 1 (id starting from
+   * 0), and size of {@code axisLabels} should be 10 as well.
+   *
+   * @param axisLabels A list of labels, whose size should be same with the size of the tensor on
+   *     the to-be-labeled axis.
+   * @param tensorBuffer The TensorBuffer to be labeled.
+   */
+  public TensorLabel(@NonNull List<String> axisLabels, @NonNull TensorBuffer tensorBuffer) {
+    this(makeMap(getFirstAxisWithSizeGreaterThanOne(tensorBuffer), axisLabels), tensorBuffer);
+  }
+
+  /**
+   * Gets the map with a pair of the label and the corresponding TensorBuffer. Only allow the
+   * mapping on the first axis with size greater than 1 currently.
+   */
+  @NonNull
+  public Map<String, TensorBuffer> getMapWithTensorBuffer() {
+    int labeledAxis = getFirstAxisWithSizeGreaterThanOne(tensorBuffer);
+
+    Map<String, TensorBuffer> labelToTensorMap = new LinkedHashMap<>();
+    SupportPreconditions.checkArgument(
+        axisLabels.containsKey(labeledAxis),
+        "get a <String, TensorBuffer> map requires the labels are set on the first non-1 axis.");
+    List<String> labels = axisLabels.get(labeledAxis);
+
+    DataType dataType = tensorBuffer.getDataType();
+    int typeSize = tensorBuffer.getTypeSize();
+    int flatSize = tensorBuffer.getFlatSize();
+
+    // Gets the underlying bytes that could be used to generate the sub-array later.
+    ByteBuffer byteBuffer = tensorBuffer.getBuffer();
+    byteBuffer.rewind();
+
+    // Note: computation below is only correct when labeledAxis is the first axis with size greater
+    // than 1.
+    int subArrayLength = flatSize / shape[labeledAxis] * typeSize;
+    int i = 0;
+    SupportPreconditions.checkNotNull(labels, "Label list should never be null");
+    for (String label : labels) {
+      // Gets the corresponding TensorBuffer.
+      byteBuffer.position(i * subArrayLength);
+      ByteBuffer subBuffer = byteBuffer.slice();
+      // ByteBuffer.slice doesn't keep order. Modify it to align with the original one.
+      subBuffer.order(byteBuffer.order()).limit(subArrayLength);
+      TensorBuffer labelBuffer = TensorBuffer.createDynamic(dataType);
+      labelBuffer.loadBuffer(subBuffer, Arrays.copyOfRange(shape, labeledAxis + 1, shape.length));
+      labelToTensorMap.put(label, labelBuffer);
+      i += 1;
+    }
+    return labelToTensorMap;
+  }
+
+  /**
+   * Gets a map that maps label to float. Only allow the mapping on the first axis with size greater
+   * than 1, and the axis should be effectively the last axis (which means every sub tensor
+   * specified by this axis should have a flat size of 1).
+   *
+   * <p>{@link TensorLabel#getCategoryList()} is an alternative API to get the result.
+   *
+   * @throws IllegalStateException if size of a sub tensor on each label is not 1.
+   */
+  @NonNull
+  public Map<String, Float> getMapWithFloatValue() {
+    int labeledAxis = getFirstAxisWithSizeGreaterThanOne(tensorBuffer);
+    SupportPreconditions.checkState(
+        labeledAxis == shape.length - 1,
+        "get a <String, Scalar> map is only valid when the only labeled axis is the last one.");
+    List<String> labels = axisLabels.get(labeledAxis);
+    float[] data = tensorBuffer.getFloatArray();
+    SupportPreconditions.checkState(labels.size() == data.length);
+    Map<String, Float> result = new LinkedHashMap<>();
+    int i = 0;
+    for (String label : labels) {
+      result.put(label, data[i]);
+      i += 1;
+    }
+    return result;
+  }
+
+  /**
+   * Gets a list of {@link Category} from the {@link TensorLabel} object.
+   *
+   * <p>The axis of label should be effectively the last axis (which means every sub tensor
+   * specified by this axis should have a flat size of 1), so that each labelled sub tensor could be
+   * converted into a float value score. Example: A {@link TensorLabel} with shape {@code {2, 5, 3}}
+   * and axis 2 is valid. If axis is 1 or 0, it cannot be converted into a {@link Category}.
+   *
+   * <p>{@link TensorLabel#getMapWithFloatValue()} is an alternative but returns a {@link Map} as
+   * the result.
+   *
+   * @throws IllegalStateException if size of a sub tensor on each label is not 1.
+   */
+  @NonNull
+  public List<Category> getCategoryList() {
+    int labeledAxis = getFirstAxisWithSizeGreaterThanOne(tensorBuffer);
+    SupportPreconditions.checkState(
+        labeledAxis == shape.length - 1,
+        "get a Category list is only valid when the only labeled axis is the last one.");
+    List<String> labels = axisLabels.get(labeledAxis);
+    float[] data = tensorBuffer.getFloatArray();
+    SupportPreconditions.checkState(labels.size() == data.length);
+    List<Category> result = new ArrayList<>();
+    int i = 0;
+    for (String label : labels) {
+      result.add(new Category(label, data[i]));
+      i += 1;
+    }
+    return result;
+  }
+
+  private static int getFirstAxisWithSizeGreaterThanOne(@NonNull TensorBuffer tensorBuffer) {
+    int[] shape = tensorBuffer.getShape();
+    for (int i = 0; i < shape.length; i++) {
+      if (shape[i] > 1) {
+        return i;
+      }
+    }
+    throw new IllegalArgumentException(
+        "Cannot find an axis to label. A valid axis to label should have size larger than 1.");
+  }
+
+  // Helper function to wrap the List<String> to a one-entry map.
+  private static Map<Integer, List<String>> makeMap(int axis, List<String> labels) {
+    Map<Integer, List<String>> map = new LinkedHashMap<>();
+    map.put(axis, labels);
+    return map;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/ops/LabelAxisOp.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/ops/LabelAxisOp.java
new file mode 100644
index 0000000..c2de8c0
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/label/ops/LabelAxisOp.java
@@ -0,0 +1,74 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.label.ops;
+
+import android.content.Context;
+import java.io.IOException;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.support.common.FileUtil;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+import org.tensorflow.lite.support.label.TensorLabel;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Labels TensorBuffer with axisLabels for outputs.
+ *
+ * <p>Apply on a {@code TensorBuffer} to get a {@code TensorLabel} that could output a Map, which is
+ * a pair of the label name and the corresponding TensorBuffer value.
+ */
+public class LabelAxisOp {
+  // Axis and its corresponding label names.
+  private final Map<Integer, List<String>> axisLabels;
+
+  protected LabelAxisOp(Builder builder) {
+    axisLabels = builder.axisLabels;
+  }
+
+  public TensorLabel apply(@NonNull TensorBuffer buffer) {
+    SupportPreconditions.checkNotNull(buffer, "Tensor buffer cannot be null.");
+    return new TensorLabel(axisLabels, buffer);
+  }
+
+  /** The inner builder class to build a LabelTensor Operator. */
+  public static class Builder {
+    private final Map<Integer, List<String>> axisLabels;
+
+    protected Builder() {
+      axisLabels = new HashMap<>();
+    }
+
+    public Builder addAxisLabel(@NonNull Context context, int axis, @NonNull String filePath)
+        throws IOException {
+      SupportPreconditions.checkNotNull(context, "Context cannot be null.");
+      SupportPreconditions.checkNotNull(filePath, "File path cannot be null.");
+      List<String> labels = FileUtil.loadLabels(context, filePath);
+      axisLabels.put(axis, labels);
+      return this;
+    }
+
+    public Builder addAxisLabel(int axis, @NonNull List<String> labels) {
+      axisLabels.put(axis, labels);
+      return this;
+    }
+
+    public LabelAxisOp build() {
+      return new LabelAxisOp(this);
+    }
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/model/GpuDelegateProxy.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/model/GpuDelegateProxy.java
new file mode 100644
index 0000000..9cfcf92
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/model/GpuDelegateProxy.java
@@ -0,0 +1,69 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.model;
+
+import android.util.Log;
+import java.io.Closeable;
+import java.io.IOException;
+import org.checkerframework.checker.nullness.qual.Nullable;
+import org.tensorflow.lite.Delegate;
+
+/**
+ * Helper class to create and call necessary methods of {@code GpuDelegate} which is not a strict
+ * dependency.
+ */
+class GpuDelegateProxy implements Delegate, Closeable {
+
+  private static final String TAG = "GpuDelegateProxy";
+
+  private final Delegate proxiedDelegate;
+  private final Closeable proxiedCloseable;
+
+  @Nullable
+  public static GpuDelegateProxy maybeNewInstance() {
+    try {
+      Class<?> clazz = Class.forName("org.tensorflow.lite.gpu.GpuDelegate");
+      Object instance = clazz.getDeclaredConstructor().newInstance();
+      return new GpuDelegateProxy(instance);
+    } catch (ReflectiveOperationException e) {
+      Log.e(TAG, "Failed to create the GpuDelegate dynamically.", e);
+      return null;
+    }
+  }
+
+  /** Calls {@code close()} method of the delegate. */
+  @Override
+  public void close() {
+    try {
+      proxiedCloseable.close();
+    } catch (IOException e) {
+      // Should not trigger, because GpuDelegate#close never throws. The catch is required because
+      // of Closeable#close.
+      Log.e(TAG, "Failed to close the GpuDelegate.", e);
+    }
+  }
+
+  /** Calls {@code getNativeHandle()} method of the delegate. */
+  @Override
+  public long getNativeHandle() {
+    return proxiedDelegate.getNativeHandle();
+  }
+
+  private GpuDelegateProxy(Object instance) {
+    this.proxiedCloseable = (Closeable) instance;
+    this.proxiedDelegate = (Delegate) instance;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/model/Model.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/model/Model.java
new file mode 100644
index 0000000..8062d68
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/model/Model.java
@@ -0,0 +1,285 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.model;
+
+import android.content.Context;
+import java.io.IOException;
+import java.nio.MappedByteBuffer;
+import java.util.Map;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.checkerframework.checker.nullness.qual.Nullable;
+import org.tensorflow.lite.Interpreter;
+import org.tensorflow.lite.Tensor;
+import org.tensorflow.lite.support.common.FileUtil;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+
+/**
+ * The wrapper class for a TFLite model and a TFLite interpreter.
+ *
+ * <p>Note: A {@link Model} can only holds 1 TFLite model at a time, and always holds a TFLite
+ * interpreter instance to run it.
+ */
+public class Model {
+
+  /** The runtime device type used for executing classification. */
+  public enum Device {
+    CPU,
+    NNAPI,
+    GPU
+  }
+
+  /**
+   * Options for running the model. Configurable parameters includes:
+   *
+   * <ul>
+   *   <li>{@code device} {@link Builder#setDevice(Device)} specifies the hardware to run the model.
+   *       The default value is {@link Device#CPU}.
+   *   <li>{@code numThreads} {@link Builder#setNumThreads(int)} specifies the number of threads
+   *       used by TFLite inference. It's only effective when device is set to {@link Device#CPU}
+   *       and default value is 1.
+   * </ul>
+   */
+  public static class Options {
+    private final Device device;
+    private final int numThreads;
+
+    /** Builder of {@link Options}. See its doc for details. */
+    public static class Builder {
+      private Device device = Device.CPU;
+      private int numThreads = 1;
+
+      public Builder setDevice(Device device) {
+        this.device = device;
+        return this;
+      }
+
+      public Builder setNumThreads(int numThreads) {
+        this.numThreads = numThreads;
+        return this;
+      }
+
+      public Options build() {
+        return new Options(this);
+      }
+    }
+
+    private Options(Builder builder) {
+      device = builder.device;
+      numThreads = builder.numThreads;
+    }
+  }
+
+  /** An instance of the driver class to run model inference with Tensorflow Lite. */
+  private final Interpreter interpreter;
+
+  /** Path to tflite model file in asset folder. */
+  private final String modelPath;
+
+  /** The memory-mapped model data. */
+  private final MappedByteBuffer byteModel;
+
+  private final GpuDelegateProxy gpuDelegateProxy;
+
+  /**
+   * Builder for {@link Model}.
+   *
+   * @deprecated Please use {@link Model#createModel(Context, String, Options)}.
+   */
+  @Deprecated
+  public static class Builder {
+    private Device device = Device.CPU;
+    private int numThreads = 1;
+    private final String modelPath;
+    private final MappedByteBuffer byteModel;
+
+    /**
+     * Creates a builder which loads tflite model from asset folder using memory-mapped files.
+     *
+     * @param context: Application context to access assets.
+     * @param modelPath: Asset path of the model (.tflite file).
+     * @throws IOException if an I/O error occurs when loading the tflite model.
+     */
+    @NonNull
+    public Builder(@NonNull Context context, @NonNull String modelPath) throws IOException {
+      this.modelPath = modelPath;
+      byteModel = FileUtil.loadMappedFile(context, modelPath);
+    }
+
+    /** Sets running device. By default, TFLite will run on CPU. */
+    @NonNull
+    public Builder setDevice(Device device) {
+      this.device = device;
+      return this;
+    }
+
+    /** Sets number of threads. By default it's 1. */
+    @NonNull
+    public Builder setNumThreads(int numThreads) {
+      this.numThreads = numThreads;
+      return this;
+    }
+
+    // Note: The implementation is copied from `Model#createModel`. As the builder is going to be
+    // deprecated, this function is also to be removed.
+    @NonNull
+    public Model build() {
+      Options options = new Options.Builder().setNumThreads(numThreads).setDevice(device).build();
+      return createModel(byteModel, modelPath, options);
+    }
+  }
+
+  /**
+   * Loads a model from assets and initialize TFLite interpreter.
+   *
+   * <p>The default options are: (1) CPU device; (2) one thread.
+   *
+   * @param context The App Context.
+   * @param modelPath The path of the model file.
+   * @throws IOException if any exception occurs when open the model file.
+   */
+  public static Model createModel(@NonNull Context context, @NonNull String modelPath)
+      throws IOException {
+    return createModel(context, modelPath, new Options.Builder().build());
+  }
+
+  /**
+   * Loads a model from assets and initialize TFLite interpreter with given options.
+   *
+   * @see Options for details.
+   * @param context The App Context.
+   * @param modelPath The path of the model file.
+   * @param options The options for running the model.
+   * @throws IOException if any exception occurs when open the model file.
+   */
+  public static Model createModel(
+      @NonNull Context context, @NonNull String modelPath, @NonNull Options options)
+      throws IOException {
+    SupportPreconditions.checkNotEmpty(
+        modelPath, "Model path in the asset folder cannot be empty.");
+    MappedByteBuffer byteModel = FileUtil.loadMappedFile(context, modelPath);
+    return createModel(byteModel, modelPath, options);
+  }
+
+  /**
+   * Creates a model with loaded {@link MappedByteBuffer}.
+   *
+   * @see Options for details.
+   * @param byteModel The loaded TFLite model.
+   * @param modelPath The original path of the model. It can be fetched later by {@link
+   *     Model#getPath()}.
+   * @param options The options for running the model.
+   * @throws IllegalArgumentException if {@code options.device} is {@link Device#GPU} but
+   *     "tensorflow-lite-gpu" is not linked to the project.
+   */
+  public static Model createModel(
+      @NonNull MappedByteBuffer byteModel, @NonNull String modelPath, @NonNull Options options) {
+    Interpreter.Options interpreterOptions = new Interpreter.Options();
+    GpuDelegateProxy gpuDelegateProxy = null;
+    switch (options.device) {
+      case NNAPI:
+        interpreterOptions.setUseNNAPI(true);
+        break;
+      case GPU:
+        gpuDelegateProxy = GpuDelegateProxy.maybeNewInstance();
+        SupportPreconditions.checkArgument(
+            gpuDelegateProxy != null,
+            "Cannot inference with GPU. Did you add \"tensorflow-lite-gpu\" as dependency?");
+        interpreterOptions.addDelegate(gpuDelegateProxy);
+        break;
+      case CPU:
+        break;
+    }
+    interpreterOptions.setNumThreads(options.numThreads);
+    Interpreter interpreter = new Interpreter(byteModel, interpreterOptions);
+    return new Model(modelPath, byteModel, interpreter, gpuDelegateProxy);
+  }
+
+  /** Returns the memory-mapped model data. */
+  @NonNull
+  public MappedByteBuffer getData() {
+    return byteModel;
+  }
+
+  /** Returns the path of the model file stored in Assets. */
+  @NonNull
+  public String getPath() {
+    return modelPath;
+  }
+
+  /**
+   * Gets the Tensor associated with the provdied input index.
+   *
+   * @throws IllegalStateException if the interpreter is closed.
+   */
+  public Tensor getInputTensor(int inputIndex) {
+    return interpreter.getInputTensor(inputIndex);
+  }
+
+  /**
+   * Gets the Tensor associated with the provdied output index.
+   *
+   * @throws IllegalStateException if the interpreter is closed.
+   */
+  public Tensor getOutputTensor(int outputIndex) {
+    return interpreter.getOutputTensor(outputIndex);
+  }
+
+  /**
+   * Returns the output shape. Useful if output shape is only determined when graph is created.
+   *
+   * @throws IllegalStateException if the interpreter is closed.
+   */
+  public int[] getOutputTensorShape(int outputIndex) {
+    return interpreter.getOutputTensor(outputIndex).shape();
+  }
+
+  /**
+   * Runs model inference on multiple inputs, and returns multiple outputs.
+   *
+   * @param inputs an array of input data. The inputs should be in the same order as inputs of the
+   *     model. Each input can be an array or multidimensional array, or a {@link
+   *     java.nio.ByteBuffer} of primitive types including int, float, long, and byte. {@link
+   *     java.nio.ByteBuffer} is the preferred way to pass large input data, whereas string types
+   *     require using the (multi-dimensional) array input path. When {@link java.nio.ByteBuffer} is
+   *     used, its content should remain unchanged until model inference is done.
+   * @param outputs a map mapping output indices to multidimensional arrays of output data or {@link
+   *     java.nio.ByteBuffer}s of primitive types including int, float, long, and byte. It only
+   *     needs to keep entries for the outputs to be used.
+   */
+  public void run(@NonNull Object[] inputs, @NonNull Map<Integer, Object> outputs) {
+    interpreter.runForMultipleInputsOutputs(inputs, outputs);
+  }
+
+  public void close() {
+    if (interpreter != null) {
+      interpreter.close();
+    }
+    if (gpuDelegateProxy != null) {
+      gpuDelegateProxy.close();
+    }
+  }
+
+  private Model(
+      @NonNull String modelPath,
+      @NonNull MappedByteBuffer byteModel,
+      @NonNull Interpreter interpreter,
+      @Nullable GpuDelegateProxy gpuDelegateProxy) {
+    this.modelPath = modelPath;
+    this.byteModel = byteModel;
+    this.interpreter = interpreter;
+    this.gpuDelegateProxy = gpuDelegateProxy;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBuffer.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBuffer.java
new file mode 100644
index 0000000..446d0ea
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBuffer.java
@@ -0,0 +1,430 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.tensorbuffer;
+
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkArgument;
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkNotNull;
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkState;
+
+import java.nio.ByteBuffer;
+import java.nio.ByteOrder;
+import java.util.Arrays;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.DataType;
+
+/** Represents the data buffer for either a model's input or its output. */
+public abstract class TensorBuffer {
+  /** Where the data is stored. */
+  protected ByteBuffer buffer;
+
+  /** Shape of the tensor stored in this buffer. */
+  protected int[] shape;
+
+  /** Number of elements in the buffer. It will be changed to a proper value in the constructor. */
+  protected int flatSize = -1;
+
+  /**
+   * Indicator of whether this buffer is dynamic or fixed-size. Fixed-size buffers will have
+   * pre-allocated memory and fixed size. While the size of dynamic buffers can be changed.
+   */
+  protected final boolean isDynamic;
+
+  /**
+   * Creates a {@link TensorBuffer} with specified {@code shape} and {@link DataType}. Here are some
+   * examples:
+   *
+   * <pre>
+   * Creating a float TensorBuffer with shape {2, 3}:
+   * int[] shape = new int[] {2, 3};
+   * TensorBuffer tensorBuffer = TensorBuffer.createFixedSize(shape, DataType.FLOAT32);
+   * </pre>
+   *
+   * <pre>
+   * Creating an uint8 TensorBuffer of a scalar:
+   * int[] shape = new int[] {};
+   * TensorBuffer tensorBuffer = TensorBuffer.createFixedSize(shape, DataType.UINT8);
+   * </pre>
+   *
+   * <pre>
+   * Creating an empty uint8 TensorBuffer:
+   * int[] shape = new int[] {0};
+   * TensorBuffer tensorBuffer = TensorBuffer.createFixedSize(shape, DataType.UINT8);
+   * </pre>
+   *
+   * <p>The size of a fixed-size TensorBuffer cannot be changed once it is created.
+   *
+   * @param shape The shape of the {@link TensorBuffer} to be created.
+   * @param dataType The dataType of the {@link TensorBuffer} to be created.
+   * @throws NullPointerException if {@code shape} is null.
+   * @throws IllegalArgumentException if {@code shape} has non-positive elements.
+   */
+  @NonNull
+  public static TensorBuffer createFixedSize(@NonNull int[] shape, DataType dataType) {
+    switch (dataType) {
+      case FLOAT32:
+        return new TensorBufferFloat(shape);
+      case UINT8:
+        return new TensorBufferUint8(shape);
+      default:
+        throw new AssertionError("TensorBuffer does not support data type: " + dataType);
+    }
+  }
+
+  /**
+   * Creates an empty dynamic {@link TensorBuffer} with specified {@link DataType}. The shape of the
+   * created {@link TensorBuffer} is {0}.
+   *
+   * <p>Dynamic TensorBuffers will reallocate memory when loading arrays or data buffers of
+   * different buffer sizes.
+   *
+   * @param dataType The dataType of the {@link TensorBuffer} to be created.
+   */
+  @NonNull
+  public static TensorBuffer createDynamic(DataType dataType) {
+    switch (dataType) {
+      case FLOAT32:
+        return new TensorBufferFloat();
+      case UINT8:
+        return new TensorBufferUint8();
+      default:
+        throw new AssertionError("TensorBuffer does not support data type: " + dataType);
+    }
+  }
+
+  /**
+   * Creates a {@link TensorBuffer} deep-copying data from another, with specified {@link DataType}.
+   *
+   * @param buffer the source {@link TensorBuffer} to copy from.
+   * @param dataType the expected {@link DataType} of newly created {@link TensorBuffer}.
+   * @throws NullPointerException if {@code buffer} is null.
+   */
+  @NonNull
+  public static TensorBuffer createFrom(@NonNull TensorBuffer buffer, DataType dataType) {
+    checkNotNull(buffer, "Cannot create a buffer from null");
+    TensorBuffer result;
+    if (buffer.isDynamic()) {
+      result = createDynamic(dataType);
+    } else {
+      result = createFixedSize(buffer.shape, dataType);
+    }
+    // The only scenario we need float array is FLOAT32->FLOAT32, or we can always use INT as
+    // intermediate container.
+    // The assumption is not true when we support other data types.
+    if (buffer.getDataType() == DataType.FLOAT32 && dataType == DataType.FLOAT32) {
+      float[] data = buffer.getFloatArray();
+      result.loadArray(data, buffer.shape);
+    } else {
+      int[] data = buffer.getIntArray();
+      result.loadArray(data, buffer.shape);
+    }
+    return result;
+  }
+
+  /** Returns the data buffer. */
+  @NonNull
+  public ByteBuffer getBuffer() {
+    return buffer;
+  }
+
+  /**
+   * Gets the {@link TensorBuffer#flatSize} of the buffer.
+   *
+   * @throws IllegalStateException if the underlying data is corrupted
+   */
+  public int getFlatSize() {
+    assertShapeIsCorect();
+    return flatSize;
+  }
+
+  /**
+   * Gets the current shape. (returning a copy here to avoid unexpected modification.)
+   *
+   * @throws IllegalStateException if the underlying data is corrupted
+   */
+  @NonNull
+  public int[] getShape() {
+    assertShapeIsCorect();
+    return Arrays.copyOf(shape, shape.length);
+  }
+
+  /** Returns the data type of this buffer. */
+  public abstract DataType getDataType();
+
+  /**
+   * Returns a float array of the values stored in this buffer. If the buffer is of different types
+   * than float, the values will be converted into float. For example, values in {@link
+   * TensorBufferUint8} will be converted from uint8 to float.
+   */
+  @NonNull
+  public abstract float[] getFloatArray();
+
+  /**
+   * Returns a float value at a given index. If the buffer is of different types than float, the
+   * value will be converted into float. For example, when reading a value from {@link
+   * TensorBufferUint8}, the value will be first read out as uint8, and then will be converted from
+   * uint8 to float.
+   *
+   * <pre>
+   * For example, a TensorBuffer with shape {2, 3} that represents the following array,
+   * [[0.0f, 1.0f, 2.0f], [3.0f, 4.0f, 5.0f]].
+   *
+   * The fourth element (whose value is 3.0f) in the TensorBuffer can be retrived by:
+   * float v = tensorBuffer.getFloatValue(3);
+   * </pre>
+   *
+   * @param absIndex The absolute index of the value to be read.
+   */
+  public abstract float getFloatValue(int absIndex);
+
+  /**
+   * Returns an int array of the values stored in this buffer. If the buffer is of different type
+   * than int, the values will be converted into int, and loss of precision may apply. For example,
+   * getting an int array from a {@link TensorBufferFloat} with values {400.32f, 23.04f}, the output
+   * is {400, 23}.
+   */
+  @NonNull
+  public abstract int[] getIntArray();
+
+  /**
+   * Returns an int value at a given index. If the buffer is of different types than int, the value
+   * will be converted into int. For example, when reading a value from {@link TensorBufferFloat},
+   * the value will be first read out as float, and then will be converted from float to int. Loss
+   * of precision may apply.
+   *
+   * <pre>
+   * For example, a TensorBuffer with shape {2, 3} that represents the following array,
+   * [[0.0f, 1.0f, 2.0f], [3.0f, 4.0f, 5.0f]].
+   *
+   * The fourth element (whose value is 3.0f) in the TensorBuffer can be retrived by:
+   * int v = tensorBuffer.getIntValue(3);
+   * Note that v is converted from 3.0f to 3 as a result of type conversion.
+   * </pre>
+   *
+   * @param absIndex The absolute index of the value to be read.
+   */
+  public abstract int getIntValue(int absIndex);
+
+  /**
+   * Returns the number of bytes of a single element in the array. For example, a float buffer will
+   * return 4, and a byte buffer will return 1.
+   */
+  public abstract int getTypeSize();
+
+  /** Returns if the {@link TensorBuffer} is dynamic sized (could resize arbitrarily). */
+  public boolean isDynamic() {
+    return isDynamic;
+  }
+
+  /**
+   * Loads an int array into this buffer with specific shape. If the buffer is of different types
+   * than int, the values will be converted into the buffer's type before being loaded into the
+   * buffer, and loss of precision may apply. For example, loading an int array with values {400,
+   * -23} into a {@link TensorBufferUint8} , the values will be clamped to [0, 255] and then be
+   * casted to uint8 by {255, 0}.
+   *
+   * @param src The source array to be loaded.
+   * @param shape Shape of the tensor that {@code src} represents.
+   * @throws NullPointerException if {@code src} is null.
+   * @throws NullPointerException if {@code shape} is null.
+   * @throws IllegalArgumentException if the size of the array to be loaded does not match the
+   *     specified shape.
+   */
+  public abstract void loadArray(@NonNull int[] src, @NonNull int[] shape);
+
+  /**
+   * Loads an int array into this buffer. If the buffer is of different types than int, the values
+   * will be converted into the buffer's type before being loaded into the buffer, and loss of
+   * precision may apply. For example, loading an int array with values {400, -23} into a {@link
+   * TensorBufferUint8} , the values will be clamped to [0, 255] and then be casted to uint8 by
+   * {255, 0}.
+   *
+   * <p>Size of {@code src} should always match the flat size of this {@link TensorBuffer}, for both
+   * fixed-size and dynamic {@link TensorBuffer}.
+   *
+   * @param src The source array to be loaded.
+   */
+  public void loadArray(@NonNull int[] src) {
+    loadArray(src, shape);
+  }
+
+  /**
+   * Loads a float array into this buffer with specific shape. If the buffer is of different types
+   * than float, the values will be converted into the buffer's type before being loaded into the
+   * buffer, and loss of precision may apply. For example, loading a float array into a {@link
+   * TensorBufferUint8} with values {400.32f, -23.04f}, the values will be clamped to [0, 255] and
+   * then be casted to uint8 by {255, 0}.
+   *
+   * @param src The source array to be loaded.
+   * @param shape Shape of the tensor that {@code src} represents.
+   * @throws NullPointerException if {@code src} is null.
+   * @throws NullPointerException if {@code shape} is null.
+   * @throws IllegalArgumentException if the size of the array to be loaded does not match the
+   *     specified shape.
+   */
+  public abstract void loadArray(@NonNull float[] src, @NonNull int[] shape);
+
+  /**
+   * Loads a float array into this buffer. If the buffer is of different types than float, the
+   * values will be converted into the buffer's type before being loaded into the buffer, and loss
+   * of precision may apply. For example, loading a float array into a {@link TensorBufferUint8}
+   * with values {400.32f, -23.04f}, the values will be clamped to [0, 255] and then be casted to
+   * uint8 by {255, 0}.
+   *
+   * <p>Size of {@code src} should always match the flat size of this {@link TensorBuffer}, for both
+   * fixed-size and dynamic {@link TensorBuffer}.
+   *
+   * @param src The source array to be loaded.
+   */
+  public void loadArray(@NonNull float[] src) {
+    loadArray(src, shape);
+  }
+
+  /**
+   * Loads a byte buffer into this {@link TensorBuffer} with specific shape.
+   *
+   * <p>Important: The loaded buffer is a reference. DO NOT MODIFY. We don't create a copy here for
+   * performance concern, but if modification is necessary, please make a copy.
+   *
+   * @param buffer The byte buffer to load.
+   * @throws NullPointerException if {@code buffer} is null.
+   * @throws IllegalArgumentException if the size of {@code buffer} and {@code typeSize} do not
+   *     match or the size of {@code buffer} and {@code flatSize} do not match.
+   */
+  public void loadBuffer(@NonNull ByteBuffer buffer, @NonNull int[] shape) {
+    checkNotNull(buffer, "Byte buffer cannot be null.");
+    int flatSize = computeFlatSize(shape);
+    checkArgument(
+        (buffer.limit() == getTypeSize() * flatSize),
+        "The size of byte buffer and the shape do not match.");
+
+    resize(shape);
+    buffer.rewind();
+    this.buffer = buffer;
+  }
+
+  /**
+   * Loads a byte buffer into this {@link TensorBuffer}. Buffer size must match the flat size of
+   * this {@link TensorBuffer}.
+   *
+   * <p>Important: The loaded buffer is a reference. DO NOT MODIFY. We don't create a copy here for
+   * performance concern, but if modification is necessary, please make a copy.
+   *
+   * @param buffer The byte buffer to load.
+   */
+  public void loadBuffer(@NonNull ByteBuffer buffer) {
+    loadBuffer(buffer, shape);
+  }
+
+  /**
+   * Constructs a fixed size {@link TensorBuffer} with specified {@code shape}.
+   *
+   * @throws NullPointerException if {@code shape} is null.
+   * @throws IllegalArgumentException if {@code shape} has non-positive elements.
+   */
+  protected TensorBuffer(@NonNull int[] shape) {
+    isDynamic = false;
+    allocateMemory(shape);
+  }
+
+  /** Constructs a dynamic {@link TensorBuffer} which can be resized. */
+  protected TensorBuffer() {
+    isDynamic = true;
+    // Initialize the dynamic TensorBuffer with an empty ByteBuffer.
+    allocateMemory(new int[] {0});
+  }
+
+  /** Calculates number of elements in the buffer. */
+  protected static int computeFlatSize(@NonNull int[] shape) {
+    checkNotNull(shape, "Shape cannot be null.");
+    int prod = 1;
+    for (int s : shape) {
+      prod = prod * s;
+    }
+    return prod;
+  }
+
+  /**
+   * For dynamic buffer, resize the memory if needed. For fixed-size buffer, check if the {@code
+   * shape} of src fits the buffer size.
+   */
+  protected void resize(@NonNull int[] shape) {
+    if (isDynamic) {
+      allocateMemory(shape);
+    } else {
+      // Make sure the new shape fits the buffer size when TensorBuffer has fixed size.
+      checkArgument(Arrays.equals(shape, this.shape));
+      this.shape = shape.clone();
+    }
+  }
+
+  /**
+   * Allocates buffer with corresponding size of the {@code shape}. If shape is an empty array, this
+   * {@link TensorBuffer} will be created as a scalar and its flatSize will be 1.
+   *
+   * @throws NullPointerException if {@code shape} is null.
+   * @throws IllegalArgumentException if {@code shape} has negative elements.
+   */
+  private void allocateMemory(@NonNull int[] shape) {
+    checkNotNull(shape, "TensorBuffer shape cannot be null.");
+    checkArgument(isShapeValid(shape), "Values in TensorBuffer shape should be non-negative.");
+
+    // Check if the new shape is the same as current shape.
+    int newFlatSize = computeFlatSize(shape);
+    this.shape = shape.clone();
+    if (flatSize == newFlatSize) {
+      return;
+    }
+
+    // Update to the new shape.
+    flatSize = newFlatSize;
+    buffer = ByteBuffer.allocateDirect(flatSize * getTypeSize());
+    buffer.order(ByteOrder.nativeOrder());
+  }
+
+  /**
+   * Verifies if the shape of the {@link TensorBuffer} matched the size of the underlying {@link
+   * ByteBuffer}.
+   */
+  private void assertShapeIsCorect() {
+    int flatSize = computeFlatSize(shape);
+    checkState(
+        (buffer.limit() == getTypeSize() * flatSize),
+        String.format(
+            "The size of underlying ByteBuffer (%d) and the shape (%s) do not match. The"
+                + " ByteBuffer may have been changed.",
+            buffer.limit(), Arrays.toString(shape)));
+  }
+
+  /**
+   * Checks if {@code shape} meets one of following two requirements: 1. Elements in {@code shape}
+   * are all non-negative numbers. 2. {@code shape} is an empty array, which corresponds to scalar.
+   */
+  private static boolean isShapeValid(@NonNull int[] shape) {
+    if (shape.length == 0) {
+      // This shape refers to a scalar.
+      return true;
+    }
+
+    // This shape refers to a multidimensional array.
+    for (int s : shape) {
+      // All elements in shape should be non-negative.
+      if (s < 0) {
+        return false;
+      }
+    }
+    return true;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBufferFloat.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBufferFloat.java
new file mode 100644
index 0000000..65bbd7d
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBufferFloat.java
@@ -0,0 +1,115 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.tensorbuffer;
+
+import java.nio.FloatBuffer;
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+
+/** Represents data buffer with float values. */
+public final class TensorBufferFloat extends TensorBuffer {
+  private static final DataType DATA_TYPE = DataType.FLOAT32;
+
+  /**
+   * Creates a {@link TensorBufferFloat} with specified {@code shape}.
+   *
+   * @throws NullPointerException if {@code shape} is null.
+   * @throws IllegalArgumentException if {@code shape} has non-positive elements.
+   */
+  TensorBufferFloat(@NonNull int[] shape) {
+    super(shape);
+  }
+
+  TensorBufferFloat() {
+    super();
+  }
+
+  @Override
+  public DataType getDataType() {
+    return DATA_TYPE;
+  }
+
+  @Override
+  @NonNull
+  public float[] getFloatArray() {
+    buffer.rewind();
+    float[] arr = new float[flatSize];
+
+    FloatBuffer floatBuffer = buffer.asFloatBuffer();
+    floatBuffer.get(arr);
+    return arr;
+  }
+
+  @Override
+  public float getFloatValue(int absIndex) {
+    return buffer.getFloat(absIndex << 2);
+  }
+
+  @Override
+  @NonNull
+  public int[] getIntArray() {
+    buffer.rewind();
+    float[] floatArr = new float[flatSize];
+    buffer.asFloatBuffer().get(floatArr);
+
+    int[] intArr = new int[flatSize];
+    for (int i = 0; i < flatSize; i++) {
+      intArr[i] = (int) floatArr[i];
+    }
+    return intArr;
+  }
+
+  @Override
+  public int getIntValue(int absIndex) {
+    return (int) buffer.getFloat(absIndex << 2);
+  }
+
+  @Override
+  public int getTypeSize() {
+    return DATA_TYPE.byteSize();
+  }
+
+  @Override
+  public void loadArray(@NonNull float[] src, @NonNull int[] shape) {
+    SupportPreconditions.checkNotNull(src, "The array to be loaded cannot be null.");
+    SupportPreconditions.checkArgument(
+        src.length == computeFlatSize(shape),
+        "The size of the array to be loaded does not match the specified shape.");
+    resize(shape);
+    buffer.rewind();
+
+    FloatBuffer floatBuffer = buffer.asFloatBuffer();
+    floatBuffer.put(src);
+  }
+
+  @Override
+  public void loadArray(@NonNull int[] src, @NonNull int[] shape) {
+    SupportPreconditions.checkNotNull(src, "The array to be loaded cannot be null.");
+    SupportPreconditions.checkArgument(
+        src.length == computeFlatSize(shape),
+        "The size of the array to be loaded does not match the specified shape.");
+    resize(shape);
+    buffer.rewind();
+
+    float[] floatArray = new float[src.length];
+    int cnt = 0;
+    for (int a : src) {
+      floatArray[cnt++] = (float) a;
+    }
+    buffer.asFloatBuffer().put(floatArray);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBufferUint8.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBufferUint8.java
new file mode 100644
index 0000000..3364194
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/support/tensorbuffer/TensorBufferUint8.java
@@ -0,0 +1,121 @@
+/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.support.tensorbuffer;
+
+import org.checkerframework.checker.nullness.qual.NonNull;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.common.SupportPreconditions;
+
+/** Represents data buffer with 8-bit unsigned integer values. */
+public final class TensorBufferUint8 extends TensorBuffer {
+  private static final DataType DATA_TYPE = DataType.UINT8;
+
+  /**
+   * Creates a {@link TensorBufferUint8} with specified {@code shape}.
+   *
+   * @throws NullPointerException if {@code shape} is null.
+   * @throws IllegalArgumentException if {@code shape} has non-positive elements.
+   */
+  TensorBufferUint8(@NonNull int[] shape) {
+    super(shape);
+  }
+
+  TensorBufferUint8() {
+    super();
+  }
+
+  @Override
+  public DataType getDataType() {
+    return DATA_TYPE;
+  }
+
+  @Override
+  @NonNull
+  public float[] getFloatArray() {
+    buffer.rewind();
+    byte[] byteArr = new byte[flatSize];
+    buffer.get(byteArr);
+
+    float[] floatArr = new float[flatSize];
+    for (int i = 0; i < flatSize; i++) {
+      floatArr[i] = (float) (byteArr[i] & 0xff);
+    }
+    return floatArr;
+  }
+
+  @Override
+  public float getFloatValue(int index) {
+    return (float) (buffer.get(index) & 0xff);
+  }
+
+  @Override
+  @NonNull
+  public int[] getIntArray() {
+    buffer.rewind();
+    byte[] byteArr = new byte[flatSize];
+    buffer.get(byteArr);
+
+    int[] intArr = new int[flatSize];
+    for (int i = 0; i < flatSize; i++) {
+      intArr[i] = byteArr[i] & 0xff;
+    }
+    return intArr;
+  }
+
+  @Override
+  public int getIntValue(int index) {
+    return buffer.get(index) & 0xff;
+  }
+
+  @Override
+  public int getTypeSize() {
+    return DATA_TYPE.byteSize();
+  }
+
+  @Override
+  public void loadArray(@NonNull float[] src, @NonNull int[] shape) {
+    SupportPreconditions.checkNotNull(src, "The array to be loaded cannot be null.");
+    SupportPreconditions.checkArgument(
+        src.length == computeFlatSize(shape),
+        "The size of the array to be loaded does not match the specified shape.");
+    resize(shape);
+    buffer.rewind();
+
+    byte[] byteArr = new byte[src.length];
+    int cnt = 0;
+    for (float a : src) {
+      byteArr[cnt++] = (byte) Math.max(Math.min(a, 255.0), 0.0);
+    }
+    buffer.put(byteArr);
+  }
+
+  @Override
+  public void loadArray(@NonNull int[] src, @NonNull int[] shape) {
+    SupportPreconditions.checkNotNull(src, "The array to be loaded cannot be null.");
+    SupportPreconditions.checkArgument(
+        src.length == computeFlatSize(shape),
+        "The size of the array to be loaded does not match the specified shape.");
+    resize(shape);
+    buffer.rewind();
+
+    byte[] byteArr = new byte[src.length];
+    int cnt = 0;
+    for (float a : src) {
+      byteArr[cnt++] = (byte) Math.max(Math.min(a, 255), 0);
+    }
+    buffer.put(byteArr);
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/BUILD
new file mode 100644
index 0000000..f82b800
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/BUILD
@@ -0,0 +1,22 @@
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load("@org_tensorflow//tensorflow/java:build_defs.bzl", "JAVACOPTS")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+android_library(
+    name = "base-task-api",
+    srcs = glob(["**/*.java"]),
+    javacopts = JAVACOPTS,
+    visibility = ["//visibility:public"],
+    deps = [
+        "@com_google_auto_value",
+    ],
+)
+
+alias(
+    name = "base_task_api",
+    actual = ":base-task-api",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/BaseTaskApi.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/BaseTaskApi.java
new file mode 100644
index 0000000..b3fe9de
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/BaseTaskApi.java
@@ -0,0 +1,91 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.core;
+
+import android.util.Log;
+import java.io.Closeable;
+
+/**
+ * Base class for Task API, provides shared logic to load/unload native libs to its C++ counterpart.
+ */
+public abstract class BaseTaskApi implements Closeable {
+  private static final String TAG = BaseTaskApi.class.getSimpleName();
+
+  /**
+   * Represents a pointer to the corresponding C++ task_api object. The nativeHandle pointer is
+   * initialized from subclasses and must be released by calling {@link #deinit} after it is no
+   * longer needed.
+   */
+  private final long nativeHandle;
+
+  /** Indicates whether the {@link #nativeHandle} pointer has been released yet. */
+  private boolean closed;
+
+  /**
+   * Constructor to initialize the JNI with a pointer from C++.
+   *
+   * @param nativeHandle a pointer referencing memory allocated in C++.
+   */
+  protected BaseTaskApi(long nativeHandle) {
+    if (nativeHandle == TaskJniUtils.INVALID_POINTER) {
+      throw new IllegalArgumentException("Failed to load C++ pointer from JNI");
+    }
+    this.nativeHandle = nativeHandle;
+  }
+
+  public boolean isClosed() {
+    return closed;
+  }
+
+  /** Release the memory allocated from C++ and deregister the library from the static holder. */
+  @Override
+  public synchronized void close() {
+    if (closed) {
+      return;
+    }
+    deinit(nativeHandle);
+    closed = true;
+  }
+
+  public long getNativeHandle() {
+    return nativeHandle;
+  }
+
+  protected void checkNotClosed() {
+    if (isClosed()) {
+      throw new IllegalStateException("Internal error: The task lib has already been closed.");
+    }
+  }
+
+  @Override
+  protected void finalize() throws Throwable {
+    try {
+      if (!closed) {
+        Log.w(TAG, "Closing an already closed native lib");
+        close();
+      }
+    } finally {
+      super.finalize();
+    }
+  }
+
+  /**
+   * Releases memory pointed by the pointer in the native layer.
+   *
+   * @param nativeHandle pointer to memory allocated
+   */
+  protected abstract void deinit(long nativeHandle);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/TaskJniUtils.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/TaskJniUtils.java
new file mode 100644
index 0000000..f5c52a0
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/TaskJniUtils.java
@@ -0,0 +1,165 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.core;
+
+import android.content.Context;
+import android.content.res.AssetFileDescriptor;
+import android.util.Log;
+import java.io.FileInputStream;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.MappedByteBuffer;
+import java.nio.channels.FileChannel;
+
+/** JNI utils for Task API. */
+public class TaskJniUtils {
+  public static final long INVALID_POINTER = 0;
+  private static final String TAG = TaskJniUtils.class.getSimpleName();
+  /** Syntax sugar to get nativeHandle from empty param list. */
+  public interface EmptyHandleProvider {
+    long createHandle();
+  }
+
+  /** Syntax sugar to get nativeHandle from an array of {@link ByteBuffer}s. */
+  public interface MultipleBuffersHandleProvider {
+    long createHandle(ByteBuffer... buffers);
+  }
+
+  /** Syntax sugar to get nativeHandle from file descriptor and options. */
+  public interface FdAndOptionsHandleProvider<T> {
+    long createHandle(
+        int fileDescriptor, long fileDescriptorLength, long fileDescriptorOffset, T options);
+  }
+
+  /**
+   * Initializes the JNI and returns C++ handle with file descriptor and options for task API.
+   *
+   * @param context the Android app context
+   * @param provider provider to get C++ handle, usually returned from native call
+   * @param libName name of C++ lib to be loaded
+   * @param filePath path of the file to be loaded
+   * @param options options to set up the task API, used by the provider
+   * @return C++ handle as long
+   * @throws IOException If model file fails to load.
+   */
+  public static <T> long createHandleFromFdAndOptions(
+      Context context,
+      final FdAndOptionsHandleProvider<T> provider,
+      String libName,
+      String filePath,
+      final T options)
+      throws IOException {
+    try (AssetFileDescriptor assetFileDescriptor = context.getAssets().openFd(filePath)) {
+      return createHandleFromLibrary(
+          new EmptyHandleProvider() {
+            @Override
+            public long createHandle() {
+              return provider.createHandle(
+                  /*fileDescriptor=*/ assetFileDescriptor.getParcelFileDescriptor().getFd(),
+                  /*fileDescriptorLength=*/ assetFileDescriptor.getLength(),
+                  /*fileDescriptorOffset=*/ assetFileDescriptor.getStartOffset(),
+                  options);
+            }
+          },
+          libName);
+    }
+  }
+
+  /**
+   * Initializes the JNI and returns C++ handle by first loading the C++ library and then invokes
+   * {@link EmptyHandleProvider#createHandle()}.
+   *
+   * @param provider provider to get C++ handle, usually returned from native call
+   * @return C++ handle as long
+   */
+  public static long createHandleFromLibrary(EmptyHandleProvider provider, String libName) {
+    tryLoadLibrary(libName);
+    try {
+      return provider.createHandle();
+    } catch (Exception e) {
+      String errorMessage = "Error getting native address of native library: " + libName;
+      Log.e(TAG, errorMessage, e);
+      throw new IllegalStateException(errorMessage, e);
+    }
+  }
+
+  /**
+   * Initializes the JNI and returns C++ handle by first loading the C++ library and then invokes
+   * {@link MultipleBuffersHandleProvider#createHandle(ByteBuffer...)}.
+   *
+   * @param context app context
+   * @param provider provider to get C++ pointer, usually returned from native call
+   * @param libName name of C++ lib to load
+   * @param filePaths file paths to load
+   * @return C++ pointer as long
+   * @throws IOException If model file fails to load.
+   */
+  public static long createHandleWithMultipleAssetFilesFromLibrary(
+      Context context,
+      final MultipleBuffersHandleProvider provider,
+      String libName,
+      String... filePaths)
+      throws IOException {
+    final MappedByteBuffer[] buffers = new MappedByteBuffer[filePaths.length];
+    for (int i = 0; i < filePaths.length; i++) {
+      buffers[i] = loadMappedFile(context, filePaths[i]);
+    }
+    return createHandleFromLibrary(
+        new EmptyHandleProvider() {
+          @Override
+          public long createHandle() {
+            return provider.createHandle(buffers);
+          }
+        },
+        libName);
+  }
+
+  /**
+   * Loads a file from the asset folder through memory mapping.
+   *
+   * @param context Application context to access assets.
+   * @param filePath Asset path of the file.
+   * @return the loaded memory mapped file.
+   * @throws IOException If model file fails to load.
+   */
+  public static MappedByteBuffer loadMappedFile(Context context, String filePath)
+      throws IOException {
+    try (AssetFileDescriptor fileDescriptor = context.getAssets().openFd(filePath);
+        FileInputStream inputStream = new FileInputStream(fileDescriptor.getFileDescriptor())) {
+      FileChannel fileChannel = inputStream.getChannel();
+      long startOffset = fileDescriptor.getStartOffset();
+      long declaredLength = fileDescriptor.getDeclaredLength();
+      return fileChannel.map(FileChannel.MapMode.READ_ONLY, startOffset, declaredLength);
+    }
+  }
+
+  private TaskJniUtils() {}
+
+  /**
+   * Try load a native library, if it's already loaded return directly.
+   *
+   * @param libName name of the lib
+   */
+  static void tryLoadLibrary(String libName) {
+    try {
+      System.loadLibrary(libName);
+    } catch (UnsatisfiedLinkError e) {
+      String errorMessage = "Error loading native library: " + libName;
+      Log.e(TAG, errorMessage, e);
+      throw new UnsatisfiedLinkError(errorMessage);
+    }
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/vision/ImageProcessingOptions.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/vision/ImageProcessingOptions.java
new file mode 100644
index 0000000..0236f2c
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core/vision/ImageProcessingOptions.java
@@ -0,0 +1,117 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.core.vision;
+
+import android.graphics.Rect;
+import com.google.auto.value.AutoValue;
+
+/**
+ * Options to configure the image processing pipeline, which operates before inference.
+ *
+ * <p>The Task Library Vision API performs image preprocessing on the input image over the region of
+ * interest, so that it fits model requirements (e.g. upright 224x224 RGB) and populate the
+ * corresponding input tensor. This is performed by (in this order):
+ *
+ * <ul>
+ *   <li>cropping the frame buffer to the region of interest (which, in most cases, just covers the
+ *       entire input image),
+ *   <li>resizing it (with bilinear interpolation, aspect-ratio *not* preserved) to the dimensions
+ *       of the model input tensor,
+ *   <li>converting it to the colorspace of the input tensor (i.e. RGB, which is the only supported
+ *       colorspace for now),
+ *   <li>rotating it according to its {@link Orientation} so that inference is performed on an
+ *       "upright" image.
+ * </ul>
+ *
+ * <p>IMPORTANT: as a consequence of cropping occurring first, the provided region of interest is
+ * expressed in the unrotated frame of reference coordinates system, i.e. in {@code [0,
+ * TensorImage.getWidth()) x [0, TensorImage.getHeight())}, which are the dimensions of the
+ * underlying image data before any orientation gets applied. If the region is out of these bounds,
+ * the inference method, such as {@link ImageClassifier#classify}, will return error.
+ */
+@AutoValue
+public abstract class ImageProcessingOptions {
+
+  /**
+   * Orientation type that follows EXIF specification.
+   *
+   * <p>The name of each enum value defines the position of the 0th row and the 0th column of the
+   * image content. See the <a href="http://jpegclub.org/exif_orientation.html">EXIF orientation
+   * documentation</a> for details.
+   */
+  public enum Orientation {
+    TOP_LEFT(0),
+    TOP_RIGHT(1),
+    BOTTOM_RIGHT(2),
+    BOTTOM_LEFT(3),
+    LEFT_TOP(4),
+    RIGHT_TOP(5),
+    RIGHT_BOTTOM(6),
+    LEFT_BOTTOM(7);
+
+    private final int value;
+
+    Orientation(int value) {
+      this.value = value;
+    }
+
+    public int getValue() {
+      return value;
+    }
+  };
+
+  private static final Rect defaultRoi = new Rect();
+  private static final Orientation DEFAULT_ORIENTATION = Orientation.TOP_LEFT;
+
+  public abstract Rect getRoi();
+
+  public abstract Orientation getOrientation();
+
+  public static Builder builder() {
+    return new AutoValue_ImageProcessingOptions.Builder()
+        .setRoi(defaultRoi)
+        .setOrientation(DEFAULT_ORIENTATION);
+  }
+
+  /** Builder for {@link ImageProcessingOptions}. */
+  @AutoValue.Builder
+  public abstract static class Builder {
+
+    /**
+     * Sets the region of interest (ROI) of the image. Defaults to the entire image.
+     *
+     * <p>Cropping according to this region of interest is prepended to the pre-processing
+     * operations.
+     */
+    public abstract Builder setRoi(Rect roi);
+
+    /**
+     * Sets the orientation of the image. Defaults to {@link Orientation#TOP_LEFT}.
+     *
+     * <p>Rotation will be applied accordingly so that inference is performed on an "upright" image.
+     */
+    public abstract Builder setOrientation(Orientation orientation);
+
+    abstract Rect getRoi();
+
+    abstract ImageProcessingOptions autoBuild();
+
+    public ImageProcessingOptions build() {
+      setRoi(new Rect(getRoi())); // Make a defensive copy, since Rect is mutable.
+      return autoBuild();
+    }
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/AndroidManifest.xml b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/AndroidManifest.xml
new file mode 100644
index 0000000..d4d1dba
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/AndroidManifest.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="org.tensorflow.lite.task.text">
+  <uses-sdk android:minSdkVersion="19" android:targetSdkVersion="29"/>
+</manifest>
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/BUILD
new file mode 100644
index 0000000..695e1be
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/BUILD
@@ -0,0 +1,37 @@
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load("@org_tensorflow//tensorflow/lite/java:aar_with_jni.bzl", "aar_with_jni")
+load("@org_tensorflow//tensorflow/java:build_defs.bzl", "JAVACOPTS")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files(["AndroidManifest.xml"])
+
+android_library(
+    name = "task_library_text",
+    srcs = [
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier:nl_classifier_src",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa:bert_question_answerer_src",
+    ],
+    javacopts = JAVACOPTS,
+    manifest = "AndroidManifest.xml",
+    visibility = ["//visibility:public"],
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/text:task_text_native",
+        "@com_google_auto_value",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text:task-library-text
+aar_with_jni(
+    name = "task-library-text",
+    android_library = ":task_library_text",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/BUILD
new file mode 100644
index 0000000..a1d78d8
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/BUILD
@@ -0,0 +1,79 @@
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load("@org_tensorflow//tensorflow/java:build_defs.bzl", "JAVACOPTS")
+load("@org_tensorflow//tensorflow/lite/java:aar_with_jni.bzl", "aar_with_jni")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+filegroup(
+    name = "nl_classifier_src",
+    srcs = glob(["**/*.java"]),
+)
+
+# Java-only target, need to be used together with a native target similar to
+# third_party/tensorflow_lite_support/java/src/native/task/text/nlclassifier:nl_classifier_native.
+# Use this target when you want to provide a MutableOpResolver with customized
+# OPs and/or a subset of BuiltInOps to reduce binary size.
+android_library(
+    name = "nl_classifier_java",
+    srcs = [
+        "NLClassifier.java",
+    ],
+    javacopts = JAVACOPTS,
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "@com_google_auto_value",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# Default target that uses BuiltInOpResolver, registers all built-in OPs.
+android_library(
+    name = "nl_classifier",
+    srcs = [
+        "NLClassifier.java",
+    ],
+    javacopts = JAVACOPTS,
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/text/nlclassifier:nl_classifier_native",
+        "@com_google_auto_value",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier:nl-classifier
+aar_with_jni(
+    name = "nl-classifier",
+    android_library = ":nl_classifier",
+)
+
+# Default target that uses BuiltInOpResolver, registers all built-in OPs.
+android_library(
+    name = "bert_nl_classifier",
+    srcs = [
+        "BertNLClassifier.java",
+    ],
+    javacopts = JAVACOPTS,
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier:bert_nl_classifier_native",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier:bert-nl-classifier
+aar_with_jni(
+    name = "bert-nl-classifier",
+    android_library = ":bert_nl_classifier",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/BertNLClassifier.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/BertNLClassifier.java
new file mode 100644
index 0000000..90bea37
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/BertNLClassifier.java
@@ -0,0 +1,142 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.text.nlclassifier;
+
+import android.content.Context;
+import android.os.ParcelFileDescriptor;
+import java.io.File;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.MappedByteBuffer;
+import java.util.List;
+import org.tensorflow.lite.support.label.Category;
+import org.tensorflow.lite.task.core.BaseTaskApi;
+import org.tensorflow.lite.task.core.TaskJniUtils;
+import org.tensorflow.lite.task.core.TaskJniUtils.EmptyHandleProvider;
+
+/**
+ * Classifier API for NLClassification tasks with Bert models, categorizes string into different
+ * classes. The API expects a Bert based TFLite model with metadata populated.
+ *
+ * <p>The metadata should contain the following information:
+ *
+ * <ul>
+ *   <li>1 input_process_unit for Wordpiece/Sentencepiece Tokenizer.
+ *   <li>3 input tensors with names "ids", "mask" and "segment_ids".
+ *   <li>1 output tensor of type float32[1, 2], with a optionally attached label file. If a label
+ *       file is attached, the file should be a plain text file with one label per line, the number
+ *       of labels should match the number of categories the model outputs.
+ * </ul>
+ */
+public class BertNLClassifier extends BaseTaskApi {
+  private static final String BERT_NL_CLASSIFIER_NATIVE_LIBNAME = "task_text_jni";
+
+  /**
+   * Constructor to initialize the JNI with a pointer from C++.
+   *
+   * @param nativeHandle a pointer referencing memory allocated in C++.
+   */
+  private BertNLClassifier(long nativeHandle) {
+    super(nativeHandle);
+  }
+
+  /**
+   * Create {@link BertNLClassifier} from a model file with metadata.
+   *
+   * @param context Android context
+   * @param pathToModel Path to the classification model.
+   * @return {@link BertNLClassifier} instance.
+   * @throws IOException If model file fails to load.
+   */
+  public static BertNLClassifier createFromFile(final Context context, final String pathToModel)
+      throws IOException {
+    return createFromBuffer(TaskJniUtils.loadMappedFile(context, pathToModel));
+  }
+
+  /**
+   * Create {@link BertNLClassifier} from a {@link File} object with metadata.
+   *
+   * @param modelFile The classification model {@link File} instance.
+   * @return {@link BertNLClassifier} instance.
+   * @throws IOException If model file fails to load.
+   */
+  public static BertNLClassifier createFromFile(File modelFile) throws IOException {
+    try (ParcelFileDescriptor descriptor =
+        ParcelFileDescriptor.open(modelFile, ParcelFileDescriptor.MODE_READ_ONLY)) {
+      return new BertNLClassifier(
+          TaskJniUtils.createHandleFromLibrary(
+              new EmptyHandleProvider() {
+                @Override
+                public long createHandle() {
+                  return initJniWithFileDescriptor(descriptor.getFd());
+                }
+              },
+              BERT_NL_CLASSIFIER_NATIVE_LIBNAME));
+    }
+  }
+
+  /**
+   * Create {@link BertNLClassifier} with a model buffer.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the model
+   * @return {@link BertNLClassifier} instance
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static BertNLClassifier createFromBuffer(final ByteBuffer modelBuffer) {
+    if (!(modelBuffer.isDirect() || modelBuffer instanceof MappedByteBuffer)) {
+      throw new IllegalArgumentException(
+          "The model buffer should be either a direct ByteBuffer or a MappedByteBuffer.");
+    }
+    return new BertNLClassifier(
+        TaskJniUtils.createHandleFromLibrary(
+            new EmptyHandleProvider() {
+              @Override
+              public long createHandle() {
+                return initJniWithByteBuffer(modelBuffer);
+              }
+            },
+            BERT_NL_CLASSIFIER_NATIVE_LIBNAME));
+  }
+
+  /**
+   * Perform classification on a string input, returns classified {@link Category}s.
+   *
+   * @param text input text to the model.
+   * @return A list of Category results.
+   */
+  public List<Category> classify(String text) {
+    return classifyNative(getNativeHandle(), text);
+  }
+
+  private static native long initJniWithByteBuffer(ByteBuffer modelBuffer);
+
+  private static native long initJniWithFileDescriptor(int fd);
+
+  private static native List<Category> classifyNative(long nativeHandle, String text);
+
+  @Override
+  protected void deinit(long nativeHandle) {
+    deinitJni(nativeHandle);
+  }
+
+  /**
+   * Native implementation to release memory pointed by the pointer.
+   *
+   * @param nativeHandle pointer to memory allocated
+   */
+  private native void deinitJni(long nativeHandle);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/NLClassifier.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/NLClassifier.java
new file mode 100644
index 0000000..2bc20d8
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/nlclassifier/NLClassifier.java
@@ -0,0 +1,257 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.text.nlclassifier;
+
+import android.content.Context;
+import android.os.ParcelFileDescriptor;
+import com.google.auto.value.AutoValue;
+import java.io.File;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.MappedByteBuffer;
+import java.util.List;
+import org.tensorflow.lite.annotations.UsedByReflection;
+import org.tensorflow.lite.support.label.Category;
+import org.tensorflow.lite.task.core.BaseTaskApi;
+import org.tensorflow.lite.task.core.TaskJniUtils;
+import org.tensorflow.lite.task.core.TaskJniUtils.EmptyHandleProvider;
+
+/**
+ * Classifier API for natural language classification tasks, categorizes string into different
+ * classes.
+ *
+ * <p>The API expects a TFLite model with the following input/output tensor:
+ *
+ * <ul>
+ *   <li>Input tensor (kTfLiteString)
+ *       <ul>
+ *         <li>input of the model, accepts a string.
+ *       </ul>
+ *   <li>Output score tensor
+ *       (kTfLiteUInt8/kTfLiteInt8/kTfLiteInt16/kTfLiteFloat32/kTfLiteFloat64/kTfLiteBool)
+ *       <ul>
+ *         <li>output scores for each class, if type is one of the Int types, dequantize it, if it
+ *             is Bool type, convert the values to 0.0 and 1.0 respectively.
+ *         <li>can have an optional associated file in metadata for labels, the file should be a
+ *             plain text file with one label per line, the number of labels should match the number
+ *             of categories the model outputs. Output label tensor: optional (kTfLiteString) -
+ *             output classname for each class, should be of the same length with scores. If this
+ *             tensor is not present, the API uses score indices as classnames. - will be ignored if
+ *             output score tensor already has an associated label file.
+ *       </ul>
+ *   <li>Optional Output label tensor (kTfLiteString/kTfLiteInt32)
+ *       <ul>
+ *         <li>output classname for each class, should be of the same length with scores. If this
+ *             tensor is not present, the API uses score indices as classnames.
+ *         <li>will be ignored if output score tensor already has an associated labe file.
+ *       </ul>
+ * </ul>
+ *
+ * <p>By default the API tries to find the input/output tensors with default configurations in
+ * {@link NLClassifierOptions}, with tensor name prioritized over tensor index. The option is
+ * configurable for different TFLite models.
+ */
+public class NLClassifier extends BaseTaskApi {
+
+  /** Options to identify input and output tensors of the model. */
+  @AutoValue
+  @UsedByReflection("nl_classifier_jni.cc")
+  public abstract static class NLClassifierOptions {
+    private static final int DEFAULT_INPUT_TENSOR_INDEX = 0;
+    private static final int DEFAULT_OUTPUT_SCORE_TENSOR_INDEX = 0;
+    // By default there is no output label tensor. The label file can be attached
+    // to the output score tensor metadata.
+    private static final int DEFAULT_OUTPUT_LABEL_TENSOR_INDEX = -1;
+    private static final String DEFAULT_INPUT_TENSOR_NAME = "INPUT";
+    private static final String DEFAULT_OUTPUT_SCORE_TENSOR_NAME = "OUTPUT_SCORE";
+    private static final String DEFAULT_OUTPUT_LABEL_TENSOR_NAME = "OUTPUT_LABEL";
+
+    @UsedByReflection("nl_classifier_jni.cc")
+    abstract int inputTensorIndex();
+
+    @UsedByReflection("nl_classifier_jni.cc")
+    abstract int outputScoreTensorIndex();
+
+    @UsedByReflection("nl_classifier_jni.cc")
+    abstract int outputLabelTensorIndex();
+
+    @UsedByReflection("nl_classifier_jni.cc")
+    abstract String inputTensorName();
+
+    @UsedByReflection("nl_classifier_jni.cc")
+    abstract String outputScoreTensorName();
+
+    @UsedByReflection("nl_classifier_jni.cc")
+    abstract String outputLabelTensorName();
+
+    public static Builder builder() {
+      return new AutoValue_NLClassifier_NLClassifierOptions.Builder()
+          .setInputTensorIndex(DEFAULT_INPUT_TENSOR_INDEX)
+          .setOutputScoreTensorIndex(DEFAULT_OUTPUT_SCORE_TENSOR_INDEX)
+          .setOutputLabelTensorIndex(DEFAULT_OUTPUT_LABEL_TENSOR_INDEX)
+          .setInputTensorName(DEFAULT_INPUT_TENSOR_NAME)
+          .setOutputScoreTensorName(DEFAULT_OUTPUT_SCORE_TENSOR_NAME)
+          .setOutputLabelTensorName(DEFAULT_OUTPUT_LABEL_TENSOR_NAME);
+    }
+
+    /** Builder for {@link NLClassifierOptions}. */
+    @AutoValue.Builder
+    public abstract static class Builder {
+      public abstract Builder setInputTensorIndex(int value);
+
+      public abstract Builder setOutputScoreTensorIndex(int value);
+
+      public abstract Builder setOutputLabelTensorIndex(int value);
+
+      public abstract Builder setInputTensorName(String value);
+
+      public abstract Builder setOutputScoreTensorName(String value);
+
+      public abstract Builder setOutputLabelTensorName(String value);
+
+      public abstract NLClassifierOptions build();
+    }
+  }
+
+  private static final String NL_CLASSIFIER_NATIVE_LIBNAME = "task_text_jni";
+
+  /**
+   * Constructor to initialize the JNI with a pointer from C++.
+   *
+   * @param nativeHandle a pointer referencing memory allocated in C++.
+   */
+  protected NLClassifier(long nativeHandle) {
+    super(nativeHandle);
+  }
+
+  /**
+   * Create {@link NLClassifier} from default {@link NLClassifierOptions}.
+   *
+   * @param context Android context.
+   * @param pathToModel Path to the classification model relative to asset dir.
+   * @return {@link NLClassifier} instance.
+   * @throws IOException If model file fails to load.
+   */
+  public static NLClassifier createFromFile(Context context, String pathToModel)
+      throws IOException {
+    return createFromFileAndOptions(context, pathToModel, NLClassifierOptions.builder().build());
+  }
+
+  /**
+   * Create {@link NLClassifier} from default {@link NLClassifierOptions}.
+   *
+   * @param modelFile The classification model {@link File} instance.
+   * @return {@link NLClassifier} instance.
+   * @throws IOException If model file fails to load.
+   */
+  public static NLClassifier createFromFile(File modelFile) throws IOException {
+    return createFromFileAndOptions(modelFile, NLClassifierOptions.builder().build());
+  }
+
+  /**
+   * Create {@link NLClassifier} from {@link NLClassifierOptions}.
+   *
+   * @param context Android context
+   * @param pathToModel Path to the classification model relative to asset dir.
+   * @param options Configurations for the model.
+   * @return {@link NLClassifier} instance.
+   * @throws IOException If model file fails to load.
+   */
+  public static NLClassifier createFromFileAndOptions(
+      Context context, String pathToModel, NLClassifierOptions options) throws IOException {
+    return createFromBufferAndOptions(TaskJniUtils.loadMappedFile(context, pathToModel), options);
+  }
+
+  /**
+   * Create {@link NLClassifier} from {@link NLClassifierOptions}.
+   *
+   * @param modelFile The classification model {@link File} instance.
+   * @param options Configurations for the model.
+   * @return {@link NLClassifier} instance.
+   * @throws IOException If model file fails to load.
+   */
+  public static NLClassifier createFromFileAndOptions(
+      File modelFile, final NLClassifierOptions options) throws IOException {
+    try (ParcelFileDescriptor descriptor =
+        ParcelFileDescriptor.open(modelFile, ParcelFileDescriptor.MODE_READ_ONLY)) {
+      return new NLClassifier(
+          TaskJniUtils.createHandleFromLibrary(
+              new EmptyHandleProvider() {
+                @Override
+                public long createHandle() {
+                  return initJniWithFileDescriptor(options, descriptor.getFd());
+                }
+              },
+              NL_CLASSIFIER_NATIVE_LIBNAME));
+    }
+  }
+
+  /**
+   * Create {@link NLClassifier} with a model {@link ByteBuffer} and {@link NLClassifierOptions}.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the
+   *     classification model
+   * @param options Configurations for the model
+   * @return {@link NLClassifier} instance
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static NLClassifier createFromBufferAndOptions(
+      final ByteBuffer modelBuffer, final NLClassifierOptions options) {
+    if (!(modelBuffer.isDirect() || modelBuffer instanceof MappedByteBuffer)) {
+      throw new IllegalArgumentException(
+          "The model buffer should be either a direct ByteBuffer or a MappedByteBuffer.");
+    }
+    return new NLClassifier(
+        TaskJniUtils.createHandleFromLibrary(
+            new EmptyHandleProvider() {
+              @Override
+              public long createHandle() {
+                return initJniWithByteBuffer(options, modelBuffer);
+              }
+            },
+            NL_CLASSIFIER_NATIVE_LIBNAME));
+  }
+
+  /**
+   * Perform classification on a string input, returns classified {@link Category}s.
+   *
+   * @param text input text to the model.
+   * @return A list of Category results.
+   */
+  public List<Category> classify(String text) {
+    return classifyNative(getNativeHandle(), text);
+  }
+
+  private static native long initJniWithByteBuffer(
+      NLClassifierOptions options, ByteBuffer modelBuffer);
+
+  private static native long initJniWithFileDescriptor(NLClassifierOptions options, int fd);
+
+  private static native List<Category> classifyNative(long nativeHandle, String text);
+
+  @Override
+  protected void deinit(long nativeHandle) {
+    deinitJni(nativeHandle);
+  }
+
+  /**
+   * Native implementation to release memory pointed by the pointer.
+   *
+   * @param nativeHandle pointer to memory allocated
+   */
+  private native void deinitJni(long nativeHandle);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/BUILD
new file mode 100644
index 0000000..3dad142
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/BUILD
@@ -0,0 +1,33 @@
+load("@org_tensorflow//tensorflow/java:build_defs.bzl", "JAVACOPTS")
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load("@org_tensorflow//tensorflow/lite/java:aar_with_jni.bzl", "aar_with_jni")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+filegroup(
+    name = "bert_question_answerer_src",
+    srcs = glob(["**/*.java"]),
+)
+
+android_library(
+    name = "bert_question_answerer",
+    srcs = glob(["*.java"]),
+    javacopts = JAVACOPTS,
+    deps = [
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/text/qa:bert_question_answerer_native",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa:bert-question-answerer
+aar_with_jni(
+    name = "bert-question-answerer",
+    android_library = ":bert_question_answerer",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/BertQuestionAnswerer.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/BertQuestionAnswerer.java
new file mode 100644
index 0000000..76f562e
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/BertQuestionAnswerer.java
@@ -0,0 +1,195 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.text.qa;
+
+import android.content.Context;
+import android.os.ParcelFileDescriptor;
+import java.io.File;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.util.List;
+import org.tensorflow.lite.task.core.BaseTaskApi;
+import org.tensorflow.lite.task.core.TaskJniUtils;
+import org.tensorflow.lite.task.core.TaskJniUtils.EmptyHandleProvider;
+import org.tensorflow.lite.task.core.TaskJniUtils.MultipleBuffersHandleProvider;
+
+/** Task API for BertQA models. */
+public class BertQuestionAnswerer extends BaseTaskApi implements QuestionAnswerer {
+  private static final String BERT_QUESTION_ANSWERER_NATIVE_LIBNAME = "task_text_jni";
+
+  private BertQuestionAnswerer(long nativeHandle) {
+    super(nativeHandle);
+  }
+
+  /**
+   * Generic API to create the QuestionAnswerer for bert models with metadata populated. The API
+   * expects a Bert based TFLite model with metadata containing the following information:
+   *
+   * <ul>
+   *   <li>input_process_units for Wordpiece/Sentencepiece Tokenizer - Wordpiece Tokenizer can be
+   *       used for a <a
+   *       href="https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1">MobileBert</a>
+   *       model, Sentencepiece Tokenizer Tokenizer can be used for an <a
+   *       href="https://tfhub.dev/tensorflow/lite-model/albert_lite_base/squadv1/1">Albert</a>
+   *       model.
+   *   <li>3 input tensors with names "ids", "mask" and "segment_ids".
+   *   <li>2 output tensors with names "end_logits" and "start_logits".
+   * </ul>
+   *
+   * @param context android context
+   * @param pathToModel file path to the model with metadata. Note: The model should not be
+   *     compressed
+   * @return {@link BertQuestionAnswerer} instance
+   * @throws IOException If model file fails to load.
+   */
+  public static BertQuestionAnswerer createFromFile(Context context, String pathToModel)
+      throws IOException {
+    return new BertQuestionAnswerer(
+        TaskJniUtils.createHandleWithMultipleAssetFilesFromLibrary(
+            context,
+            new MultipleBuffersHandleProvider() {
+              @Override
+              public long createHandle(ByteBuffer... buffers) {
+                return BertQuestionAnswerer.initJniWithModelWithMetadataByteBuffers(buffers);
+              }
+            },
+            BERT_QUESTION_ANSWERER_NATIVE_LIBNAME,
+            pathToModel));
+  }
+
+  /**
+   * Generic API to create the QuestionAnswerer for bert models with metadata populated. The API
+   * expects a Bert based TFLite model with metadata containing the following information:
+   *
+   * <ul>
+   *   <li>input_process_units for Wordpiece/Sentencepiece Tokenizer - Wordpiece Tokenizer can be
+   *       used for a <a
+   *       href="https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1">MobileBert</a>
+   *       model, Sentencepiece Tokenizer Tokenizer can be used for an <a
+   *       href="https://tfhub.dev/tensorflow/lite-model/albert_lite_base/squadv1/1">Albert</a>
+   *       model.
+   *   <li>3 input tensors with names "ids", "mask" and "segment_ids".
+   *   <li>2 output tensors with names "end_logits" and "start_logits".
+   * </ul>
+   *
+   * @param modelFile {@link File} object of the model
+   * @return {@link BertQuestionAnswerer} instance
+   * @throws IOException If model file fails to load.
+   */
+  public static BertQuestionAnswerer createFromFile(File modelFile)
+      throws IOException {
+    try (ParcelFileDescriptor descriptor =
+        ParcelFileDescriptor.open(modelFile, ParcelFileDescriptor.MODE_READ_ONLY)) {
+      return new BertQuestionAnswerer(
+          TaskJniUtils.createHandleFromLibrary(
+              new EmptyHandleProvider() {
+                @Override
+                public long createHandle() {
+                  return initJniWithFileDescriptor(descriptor.getFd());
+                }
+              },
+              BERT_QUESTION_ANSWERER_NATIVE_LIBNAME));
+    }
+  }
+
+  /**
+   * Creates the API instance with a bert model and vocabulary file.
+   *
+   * <p>One suitable model is: https://tfhub.dev/tensorflow/lite-model/mobilebert/1/default/1
+   *
+   * @param context android context
+   * @param pathToModel file path to the bert model. Note: The model should not be compressed
+   * @param pathToVocab file path to the vocabulary file. Note: The file should not be compressed
+   * @return {@link BertQuestionAnswerer} instance
+   * @throws IOException If model file fails to load.
+   */
+  public static BertQuestionAnswerer createBertQuestionAnswererFromFile(
+      Context context, String pathToModel, String pathToVocab) throws IOException {
+    return new BertQuestionAnswerer(
+        TaskJniUtils.createHandleWithMultipleAssetFilesFromLibrary(
+            context,
+            new MultipleBuffersHandleProvider() {
+              @Override
+              public long createHandle(ByteBuffer... buffers) {
+                return BertQuestionAnswerer.initJniWithBertByteBuffers(buffers);
+              }
+            },
+            BERT_QUESTION_ANSWERER_NATIVE_LIBNAME,
+            pathToModel,
+            pathToVocab));
+  }
+
+  /**
+   * Creates the API instance with an albert model and sentence piece model file.
+   *
+   * <p>One suitable model is: https://tfhub.dev/tensorflow/lite-model/albert_lite_base/squadv1/1
+   *
+   * @param context android context
+   * @param pathToModel file path to the albert model. Note: The model should not be compressed
+   * @param pathToSentencePieceModel file path to the sentence piece model file. Note: The model
+   *     should not be compressed
+   * @return {@link BertQuestionAnswerer} instance
+   * @throws IOException If model file fails to load.
+   */
+  public static BertQuestionAnswerer createAlbertQuestionAnswererFromFile(
+      Context context, String pathToModel, String pathToSentencePieceModel) throws IOException {
+    return new BertQuestionAnswerer(
+        TaskJniUtils.createHandleWithMultipleAssetFilesFromLibrary(
+            context,
+            new MultipleBuffersHandleProvider() {
+              @Override
+              public long createHandle(ByteBuffer... buffers) {
+                return BertQuestionAnswerer.initJniWithAlbertByteBuffers(buffers);
+              }
+            },
+            BERT_QUESTION_ANSWERER_NATIVE_LIBNAME,
+            pathToModel,
+            pathToSentencePieceModel));
+  }
+
+  @Override
+  public List<QaAnswer> answer(String context, String question) {
+    checkNotClosed();
+    return answerNative(getNativeHandle(), context, question);
+  }
+
+  // modelBuffers[0] is tflite model file buffer, and modelBuffers[1] is vocab file buffer.
+  private static native long initJniWithBertByteBuffers(ByteBuffer... modelBuffers);
+
+  // modelBuffers[0] is tflite model file buffer, and modelBuffers[1] is sentencepiece model file
+  // buffer.
+  private static native long initJniWithAlbertByteBuffers(ByteBuffer... modelBuffers);
+
+  // modelBuffers[0] is tflite model file buffer with metadata to specify which tokenizer to use.
+  private static native long initJniWithModelWithMetadataByteBuffers(ByteBuffer... modelBuffers);
+
+  private static native long initJniWithFileDescriptor(int fd);
+
+  private static native List<QaAnswer> answerNative(
+      long nativeHandle, String context, String question);
+
+  @Override
+  protected void deinit(long nativeHandle) {
+    deinitJni(nativeHandle);
+  }
+
+  /**
+   * Native implementation to release memory pointed by the pointer.
+   *
+   * @param nativeHandle pointer to memory allocated
+   */
+  private native void deinitJni(long nativeHandle);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/QaAnswer.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/QaAnswer.java
new file mode 100644
index 0000000..4259a69
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/QaAnswer.java
@@ -0,0 +1,58 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.text.qa;
+
+import org.tensorflow.lite.annotations.UsedByReflection;
+
+/**
+ * Answers to {@link QuestionAnswerer}. Contains information about the answer and its relative
+ * position information to the context.
+ */
+public class QaAnswer {
+  public Pos pos;
+  public String text;
+
+  @UsedByReflection("bert_question_answerer_jni.cc")
+  public QaAnswer(String text, Pos pos) {
+    this.text = text;
+    this.pos = pos;
+  }
+
+  public QaAnswer(String text, int start, int end, float logit) {
+    this(text, new Pos(start, end, logit));
+  }
+
+  /**
+   * Position information of the answer relative to context. It is sortable in descending order
+   * based on logit.
+   */
+  public static class Pos implements Comparable<Pos> {
+    public int start;
+    public int end;
+    public float logit;
+
+    public Pos(int start, int end, float logit) {
+      this.start = start;
+      this.end = end;
+      this.logit = logit;
+    }
+
+    @Override
+    public int compareTo(Pos other) {
+      return Float.compare(other.logit, this.logit);
+    }
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/QuestionAnswerer.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/QuestionAnswerer.java
new file mode 100644
index 0000000..8df6d37
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text/qa/QuestionAnswerer.java
@@ -0,0 +1,32 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.text.qa;
+
+import java.util.List;
+
+/** API to answer questions based on context. */
+public interface QuestionAnswerer {
+
+  /**
+   * Answers question based on context, and returns a list of possible {@link QaAnswer}s. Could be
+   * empty if no answer was found from the given context.
+   *
+   * @param context context the question bases on
+   * @param question question to ask
+   * @return a list of possible answers in {@link QaAnswer}
+   */
+  List<QaAnswer> answer(String context, String question);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/AndroidManifest.xml b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/AndroidManifest.xml
new file mode 100644
index 0000000..e77a073
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/AndroidManifest.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="org.tensorflow.lite.task.vision">
+  <uses-sdk android:minSdkVersion="19" android:targetSdkVersion="29"/>
+</manifest>
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/BUILD
new file mode 100644
index 0000000..661a766
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/BUILD
@@ -0,0 +1,41 @@
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load("@org_tensorflow//tensorflow/lite/java:aar_with_jni.bzl", "aar_with_jni")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "AndroidManifest.xml",
+])
+
+android_library(
+    name = "task_library_vision",
+    srcs = [
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier:image_classifier_src",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector:object_detector_src",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter:image_segmenter_src",
+    ],
+    # TODO(b/163039980): Use JAVACOPTS in TF. "-Xep:RemoveUnusedImports:ERROR" wierdly break the build.
+    javacopts = ["-source 7 -target 7"],
+    manifest = "AndroidManifest.xml",
+    visibility = ["//visibility:public"],
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/vision:task_vision_native",
+        "@com_google_auto_value",
+        "@maven//:androidx_annotation_annotation",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision:task-library-vision
+aar_with_jni(
+    name = "task-library-vision",
+    android_library = ":task_library_vision",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/AndroidManifest.xml b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/AndroidManifest.xml
new file mode 100644
index 0000000..ce07182
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/AndroidManifest.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="org.tensorflow.lite.task.vision.classifier">
+  <uses-sdk android:minSdkVersion="19" android:targetSdkVersion="29"/>
+</manifest>
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/BUILD
new file mode 100644
index 0000000..c6a70a0
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/BUILD
@@ -0,0 +1,40 @@
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load("@org_tensorflow//tensorflow/lite/java:aar_with_jni.bzl", "aar_with_jni")
+load("@org_tensorflow//tensorflow/java:build_defs.bzl", "JAVACOPTS")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "AndroidManifest.xml",
+])
+
+filegroup(
+    name = "image_classifier_src",
+    srcs = glob(["**/*.java"]),
+)
+
+android_library(
+    name = "image_classifier",
+    srcs = glob(["*.java"]),
+    javacopts = JAVACOPTS,
+    manifest = "AndroidManifest.xml",
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/vision/classifier:image_classifier_native",
+        "@com_google_auto_value",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier:image-classifier
+aar_with_jni(
+    name = "image-classifier",
+    android_library = ":image_classifier",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/Classifications.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/Classifications.java
new file mode 100644
index 0000000..d33f0fb
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/Classifications.java
@@ -0,0 +1,46 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.classifier;
+
+import com.google.auto.value.AutoValue;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+import org.tensorflow.lite.annotations.UsedByReflection;
+import org.tensorflow.lite.support.label.Category;
+
+/**
+ * The classification results of one head in a multihead (a.k.a. multi-output) {@link
+ * ImageClassifier}. A multihead {@link ImageClassifier} can perform classification for multiple
+ * purposes, such as a fine grained classifier to describe apparel items (e.g. color, material,
+ * type, etc.).
+ */
+@AutoValue
+@UsedByReflection("image_classifier_jni.cc")
+public abstract class Classifications {
+
+  @UsedByReflection("image_classifier_jni.cc")
+  static Classifications create(List<Category> categories, int headIndex) {
+    return new AutoValue_Classifications(
+        Collections.unmodifiableList(new ArrayList<Category>(categories)), headIndex);
+  }
+
+  // Same reason for not using ImmutableList as stated in
+  // {@link ImageClassifier#ImageClassifierOptions#labelAllowList}.
+  public abstract List<Category> getCategories();
+
+  public abstract int getHeadIndex();
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/ImageClassifier.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/ImageClassifier.java
new file mode 100644
index 0000000..46f6754
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/classifier/ImageClassifier.java
@@ -0,0 +1,453 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.classifier;
+
+import android.content.Context;
+import android.graphics.Rect;
+import android.os.ParcelFileDescriptor;
+import java.io.File;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.MappedByteBuffer;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.annotations.UsedByReflection;
+import org.tensorflow.lite.support.image.TensorImage;
+import org.tensorflow.lite.task.core.BaseTaskApi;
+import org.tensorflow.lite.task.core.TaskJniUtils;
+import org.tensorflow.lite.task.core.TaskJniUtils.EmptyHandleProvider;
+import org.tensorflow.lite.task.core.TaskJniUtils.FdAndOptionsHandleProvider;
+import org.tensorflow.lite.task.core.vision.ImageProcessingOptions;
+
+/**
+ * Performs classification on images.
+ *
+ * <p>The API expects a TFLite model with optional, but strongly recommended, <a
+ * href="https://www.tensorflow.org/lite/convert/metadata">TFLite Model Metadata.</a>.
+ *
+ * <p>The API supports models with one image input tensor and one classification output tensor. To
+ * be more specific, here are the requirements.
+ *
+ * <ul>
+ *   <li>Input image tensor ({@code kTfLiteUInt8}/{@code kTfLiteFloat32})
+ *       <ul>
+ *         <li>image input of size {@code [batch x height x width x channels]}.
+ *         <li>batch inference is not supported ({@code batch} is required to be 1).
+ *         <li>only RGB inputs are supported ({@code channels} is required to be 3).
+ *         <li>if type is {@code kTfLiteFloat32}, NormalizationOptions are required to be attached
+ *             to the metadata for input normalization.
+ *       </ul>
+ *   <li>Output score tensor ({@code kTfLiteUInt8}/{@code kTfLiteFloat32})
+ *       <ul>
+ *         <li>with {@code N} classes of either 2 or 4 dimensions, such as {@code [1 x N]} or {@code
+ *             [1 x 1 x 1 x N]}
+ *         <li>the label file is required to be packed to the metadata. See the <a
+ *             href="https://www.tensorflow.org/lite/convert/metadata#label_output">example of
+ *             creating metadata for an image classifier</a>. If no label files are packed, it will
+ *             use index as label in the result.
+ *       </ul>
+ * </ul>
+ *
+ * <p>An example of such model can be found on <a
+ * href="https://tfhub.dev/bohemian-visual-recognition-alliance/lite-model/models/mushroom-identification_v1/1">TensorFlow
+ * Hub.</a>.
+ */
+public final class ImageClassifier extends BaseTaskApi {
+
+  private static final String IMAGE_CLASSIFIER_NATIVE_LIB = "task_vision_jni";
+  private static final int OPTIONAL_FD_LENGTH = -1;
+  private static final int OPTIONAL_FD_OFFSET = -1;
+
+  /**
+   * Creates an {@link ImageClassifier} instance from the default {@link ImageClassifierOptions}.
+   *
+   * @param modelPath path of the classification model with metadata in the assets
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageClassifier} from the native
+   *     code
+   */
+  public static ImageClassifier createFromFile(Context context, String modelPath)
+      throws IOException {
+    return createFromFileAndOptions(context, modelPath, ImageClassifierOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ImageClassifier} instance from the default {@link ImageClassifierOptions}.
+   *
+   * @param modelFile the classification model {@link File} instance
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageClassifier} from the native
+   *     code
+   */
+  public static ImageClassifier createFromFile(File modelFile) throws IOException {
+    return createFromFileAndOptions(modelFile, ImageClassifierOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ImageClassifier} instance with a model buffer and the default {@link
+   * ImageClassifierOptions}.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the
+   *     classification model
+   * @throws AssertionError if error occurs when creating {@link ImageClassifier} from the native
+   *     code
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static ImageClassifier createFromBuffer(final ByteBuffer modelBuffer) {
+    return createFromBufferAndOptions(modelBuffer, ImageClassifierOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ImageClassifier} instance from {@link ImageClassifierOptions}.
+   *
+   * @param modelPath path of the classification model with metadata in the assets
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageClassifier} from the native
+   *     code
+   */
+  public static ImageClassifier createFromFileAndOptions(
+      Context context, String modelPath, ImageClassifierOptions options) throws IOException {
+    return new ImageClassifier(
+        TaskJniUtils.createHandleFromFdAndOptions(
+            context,
+            new FdAndOptionsHandleProvider<ImageClassifierOptions>() {
+              @Override
+              public long createHandle(
+                  int fileDescriptor,
+                  long fileDescriptorLength,
+                  long fileDescriptorOffset,
+                  ImageClassifierOptions options) {
+                return initJniWithModelFdAndOptions(
+                    fileDescriptor, fileDescriptorLength, fileDescriptorOffset, options);
+              }
+            },
+            IMAGE_CLASSIFIER_NATIVE_LIB,
+            modelPath,
+            options));
+  }
+
+  /**
+   * Creates an {@link ImageClassifier} instance.
+   *
+   * @param modelFile the classification model {@link File} instance
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageClassifier} from the native
+   *     code
+   */
+  public static ImageClassifier createFromFileAndOptions(
+      File modelFile, final ImageClassifierOptions options) throws IOException {
+    try (ParcelFileDescriptor descriptor =
+        ParcelFileDescriptor.open(modelFile, ParcelFileDescriptor.MODE_READ_ONLY)) {
+      return new ImageClassifier(
+          TaskJniUtils.createHandleFromLibrary(
+              new TaskJniUtils.EmptyHandleProvider() {
+                @Override
+                public long createHandle() {
+                  return initJniWithModelFdAndOptions(
+                      descriptor.getFd(),
+                      /*fileDescriptorLength=*/ OPTIONAL_FD_LENGTH,
+                      /*fileDescriptorOffset=*/ OPTIONAL_FD_OFFSET,
+                      options);
+                }
+              },
+              IMAGE_CLASSIFIER_NATIVE_LIB));
+    }
+  }
+
+  /**
+   * Creates an {@link ImageClassifier} instance with a model buffer and {@link
+   * ImageClassifierOptions}.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the
+   *     classification model
+   * @throws AssertionError if error occurs when creating {@link ImageClassifier} from the native
+   *     code
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static ImageClassifier createFromBufferAndOptions(
+      final ByteBuffer modelBuffer, final ImageClassifierOptions options) {
+    if (!(modelBuffer.isDirect() || modelBuffer instanceof MappedByteBuffer)) {
+      throw new IllegalArgumentException(
+          "The model buffer should be either a direct ByteBuffer or a MappedByteBuffer.");
+    }
+    return new ImageClassifier(
+        TaskJniUtils.createHandleFromLibrary(
+            new EmptyHandleProvider() {
+              @Override
+              public long createHandle() {
+                return initJniWithByteBuffer(modelBuffer, options);
+              }
+            },
+            IMAGE_CLASSIFIER_NATIVE_LIB));
+  }
+
+  /**
+   * Constructor to initialize the JNI with a pointer from C++.
+   *
+   * @param nativeHandle a pointer referencing memory allocated in C++
+   */
+  private ImageClassifier(long nativeHandle) {
+    super(nativeHandle);
+  }
+
+  /** Options for setting up an ImageClassifier. */
+  @UsedByReflection("image_classifier_jni.cc")
+  public static class ImageClassifierOptions {
+    // Not using AutoValue for this class because scoreThreshold cannot have default value
+    // (otherwise, the default value would override the one in the model metadata) and `Optional` is
+    // not an option here, because
+    // 1. java.util.Optional require Java 8 while we need to support Java 7.
+    // 2. The Guava library (com.google.common.base.Optional) is avoided in this project. See the
+    // comments for labelAllowList.
+    private final String displayNamesLocale;
+    private final int maxResults;
+    private final float scoreThreshold;
+    private final boolean isScoreThresholdSet;
+    // As an open source project, we've been trying avoiding depending on common java libraries,
+    // such as Guava, because it may introduce conflicts with clients who also happen to use those
+    // libraries. Therefore, instead of using ImmutableList here, we convert the List into
+    // unmodifiableList in setLabelAllowList() and setLabelDenyList() to make it less
+    // vulnerable.
+    private final List<String> labelAllowList;
+    private final List<String> labelDenyList;
+    private final int numThreads;
+
+    public static Builder builder() {
+      return new Builder();
+    }
+
+    /** A builder that helps to configure an instance of ImageClassifierOptions. */
+    public static class Builder {
+      private String displayNamesLocale = "en";
+      private int maxResults = -1;
+      private float scoreThreshold;
+      private boolean isScoreThresholdSet = false;
+      private List<String> labelAllowList = new ArrayList<>();
+      private List<String> labelDenyList = new ArrayList<>();
+      private int numThreads = -1;
+
+      private Builder() {}
+
+      /**
+       * Sets the locale to use for display names specified through the TFLite Model Metadata, if
+       * any.
+       *
+       * <p>Defaults to English({@code "en"}). See the <a
+       * href="https://github.com/tensorflow/tflite-support/blob/3ce83f0cfe2c68fecf83e019f2acc354aaba471f/tensorflow_lite_support/metadata/metadata_schema.fbs#L147">TFLite
+       * Metadata schema file.</a> for the accepted pattern of locale.
+       */
+      public Builder setDisplayNamesLocale(String displayNamesLocale) {
+        this.displayNamesLocale = displayNamesLocale;
+        return this;
+      }
+
+      /**
+       * Sets the maximum number of top scored results to return.
+       *
+       * <p>If < 0, all results will be returned. If 0, an invalid argument error is returned.
+       * Defaults to -1.
+       *
+       * @throws IllegalArgumentException if maxResults is 0.
+       */
+      public Builder setMaxResults(int maxResults) {
+        if (maxResults == 0) {
+          throw new IllegalArgumentException("maxResults cannot be 0.");
+        }
+        this.maxResults = maxResults;
+        return this;
+      }
+
+      /**
+       * Sets the score threshold in [0,1).
+       *
+       * <p>It overrides the one provided in the model metadata (if any). Results below this value
+       * are rejected.
+       */
+      public Builder setScoreThreshold(float scoreThreshold) {
+        this.scoreThreshold = scoreThreshold;
+        isScoreThresholdSet = true;
+        return this;
+      }
+
+      /**
+       * Sets the optional allowlist of labels.
+       *
+       * <p>If non-empty, classifications whose label is not in this set will be filtered out.
+       * Duplicate or unknown labels are ignored. Mutually exclusive with labelDenyList.
+       */
+      public Builder setLabelAllowList(List<String> labelAllowList) {
+        this.labelAllowList = Collections.unmodifiableList(new ArrayList<>(labelAllowList));
+        return this;
+      }
+
+      /**
+       * Sets the optional denylist of labels.
+       *
+       * <p>If non-empty, classifications whose label is in this set will be filtered out. Duplicate
+       * or unknown labels are ignored. Mutually exclusive with labelAllowList.
+       */
+      public Builder setLabelDenyList(List<String> labelDenyList) {
+        this.labelDenyList = Collections.unmodifiableList(new ArrayList<>(labelDenyList));
+        return this;
+      }
+
+      /**
+       * Sets the number of threads to be used for TFLite ops that support multi-threading when
+       * running inference with CPU. Defaults to -1.
+       *
+       * <p>numThreads should be greater than 0 or equal to -1. Setting numThreads to -1 has the
+       * effect to let TFLite runtime set the value.
+       */
+      public Builder setNumThreads(int numThreads) {
+        this.numThreads = numThreads;
+        return this;
+      }
+
+      public ImageClassifierOptions build() {
+        return new ImageClassifierOptions(this);
+      }
+    }
+
+    @UsedByReflection("image_classifier_jni.cc")
+    public String getDisplayNamesLocale() {
+      return displayNamesLocale;
+    }
+
+    @UsedByReflection("image_classifier_jni.cc")
+    public int getMaxResults() {
+      return maxResults;
+    }
+
+    @UsedByReflection("image_classifier_jni.cc")
+    public float getScoreThreshold() {
+      return scoreThreshold;
+    }
+
+    @UsedByReflection("image_classifier_jni.cc")
+    public boolean getIsScoreThresholdSet() {
+      return isScoreThresholdSet;
+    }
+
+    @UsedByReflection("image_classifier_jni.cc")
+    public List<String> getLabelAllowList() {
+      return new ArrayList<>(labelAllowList);
+    }
+
+    @UsedByReflection("image_classifier_jni.cc")
+    public List<String> getLabelDenyList() {
+      return new ArrayList<>(labelDenyList);
+    }
+
+    @UsedByReflection("image_classifier_jni.cc")
+    public int getNumThreads() {
+      return numThreads;
+    }
+
+    private ImageClassifierOptions(Builder builder) {
+      displayNamesLocale = builder.displayNamesLocale;
+      maxResults = builder.maxResults;
+      scoreThreshold = builder.scoreThreshold;
+      isScoreThresholdSet = builder.isScoreThresholdSet;
+      labelAllowList = builder.labelAllowList;
+      labelDenyList = builder.labelDenyList;
+      numThreads = builder.numThreads;
+    }
+  }
+
+  /**
+   * Performs actual classification on the provided image.
+   *
+   * @param image a {@link TensorImage} object that represents an RGB image
+   * @throws AssertionError if error occurs when classifying the image from the native code
+   */
+  public List<Classifications> classify(TensorImage image) {
+    return classify(image, ImageProcessingOptions.builder().build());
+  }
+
+  /**
+   * Performs actual classification on the provided image with {@link ImageProcessingOptions}.
+   *
+   * <p>{@link ImageClassifier} supports the following options:
+   *
+   * <ul>
+   *   <li>Region of interest (ROI) (through {@link ImageProcessingOptions#Builder#setRoi}). It
+   *       defaults to the entire image.
+   *   <li>image rotation (through {@link ImageProcessingOptions#Builder#setOrientation}). It
+   *       defaults to {@link ImageProcessingOptions#Orientation#TOP_LEFT}.
+   * </ul>
+   *
+   * @param image a {@link TensorImage} object that represents an RGB image
+   * @throws AssertionError if error occurs when classifying the image from the native code
+   */
+  public List<Classifications> classify(TensorImage image, ImageProcessingOptions options) {
+    checkNotClosed();
+
+    // image_classifier_jni.cc expects an uint8 image. Convert image of other types into uint8.
+    TensorImage imageUint8 =
+        image.getDataType() == DataType.UINT8
+            ? image
+            : TensorImage.createFrom(image, DataType.UINT8);
+
+    Rect roi =
+        options.getRoi().isEmpty()
+            ? new Rect(0, 0, imageUint8.getWidth(), imageUint8.getHeight())
+            : options.getRoi();
+
+    return classifyNative(
+        getNativeHandle(),
+        imageUint8.getBuffer(),
+        imageUint8.getWidth(),
+        imageUint8.getHeight(),
+        new int[] {roi.left, roi.top, roi.width(), roi.height()},
+        options.getOrientation().getValue());
+  }
+
+  private static native long initJniWithModelFdAndOptions(
+      int fileDescriptor,
+      long fileDescriptorLength,
+      long fileDescriptorOffset,
+      ImageClassifierOptions options);
+
+  private static native long initJniWithByteBuffer(
+      ByteBuffer modelBuffer, ImageClassifierOptions options);
+
+  /**
+   * The native method to classify an image with the ROI and orientation.
+   *
+   * @param roi the ROI of the input image, an array representing the bounding box as {left, top,
+   *     width, height}
+   * @param orientation the integer value corresponding to {@link
+   *     ImageProcessingOptions#Orientation}
+   */
+  private static native List<Classifications> classifyNative(
+      long nativeHandle, ByteBuffer image, int width, int height, int[] roi, int orientation);
+
+  @Override
+  protected void deinit(long nativeHandle) {
+    deinitJni(nativeHandle);
+  }
+
+  /**
+   * Native implementation to release memory pointed by the pointer.
+   *
+   * @param nativeHandle pointer to memory allocated
+   */
+  private native void deinitJni(long nativeHandle);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/AndroidManifest.xml b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/AndroidManifest.xml
new file mode 100644
index 0000000..5fefccd
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/AndroidManifest.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="org.tensorflow.lite.task.vision.detector">
+  <uses-sdk android:minSdkVersion="19" android:targetSdkVersion="29"/>
+</manifest>
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/BUILD
new file mode 100644
index 0000000..d0d541a
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/BUILD
@@ -0,0 +1,40 @@
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+load("@org_tensorflow//tensorflow/lite/java:aar_with_jni.bzl", "aar_with_jni")
+load("@org_tensorflow//tensorflow/java:build_defs.bzl", "JAVACOPTS")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "AndroidManifest.xml",
+])
+
+filegroup(
+    name = "object_detector_src",
+    srcs = glob(["**/*.java"]),
+)
+
+android_library(
+    name = "object_detector",
+    srcs = glob(["*.java"]),
+    javacopts = JAVACOPTS,
+    manifest = "AndroidManifest.xml",
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/vision/detector:object_detector_native",
+        "@com_google_auto_value",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector:object-detector
+aar_with_jni(
+    name = "object-detector",
+    android_library = ":object_detector",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/Detection.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/Detection.java
new file mode 100644
index 0000000..007e032
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/Detection.java
@@ -0,0 +1,42 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.detector;
+
+import android.graphics.RectF;
+import com.google.auto.value.AutoValue;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+import org.tensorflow.lite.annotations.UsedByReflection;
+import org.tensorflow.lite.support.label.Category;
+
+/** Represents one detected object in the results of a {@link ObjectDetector}. */
+@AutoValue
+@UsedByReflection("object_detection_jni.cc")
+public abstract class Detection {
+
+  @UsedByReflection("object_detection_jni.cc")
+  public static Detection create(RectF boundingBox, List<Category> categories) {
+    return new AutoValue_Detection(
+        new RectF(boundingBox), Collections.unmodifiableList(new ArrayList<Category>(categories)));
+  }
+
+  public abstract RectF getBoundingBox();
+
+  // Same reason for not using ImmutableList as stated in
+  // {@link ObjectDetector#ObjectDetectorOptions#labelAllowList}.
+  public abstract List<Category> getCategories();
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/ObjectDetector.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/ObjectDetector.java
new file mode 100644
index 0000000..75bc983
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/detector/ObjectDetector.java
@@ -0,0 +1,452 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.detector;
+
+import android.content.Context;
+import android.os.ParcelFileDescriptor;
+import java.io.File;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.MappedByteBuffer;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.annotations.UsedByReflection;
+import org.tensorflow.lite.support.image.TensorImage;
+import org.tensorflow.lite.task.core.BaseTaskApi;
+import org.tensorflow.lite.task.core.TaskJniUtils;
+import org.tensorflow.lite.task.core.TaskJniUtils.EmptyHandleProvider;
+import org.tensorflow.lite.task.core.TaskJniUtils.FdAndOptionsHandleProvider;
+import org.tensorflow.lite.task.core.vision.ImageProcessingOptions;
+
+/**
+ * Performs object detection on images.
+ *
+ * <p>The API expects a TFLite model with <a
+ * href="https://www.tensorflow.org/lite/convert/metadata">TFLite Model Metadata.</a>.
+ *
+ * <p>The API supports models with one image input tensor and four output tensors. To be more
+ * specific, here are the requirements.
+ *
+ * <ul>
+ *   <li>Input image tensor ({@code kTfLiteUInt8}/{@code kTfLiteFloat32})
+ *       <ul>
+ *         <li>image input of size {@code [batch x height x width x channels]}.
+ *         <li>batch inference is not supported ({@code batch} is required to be 1).
+ *         <li>only RGB inputs are supported ({@code channels} is required to be 3).
+ *         <li>if type is {@code kTfLiteFloat32}, NormalizationOptions are required to be attached
+ *             to the metadata for input normalization.
+ *       </ul>
+ *   <li>Output tensors must be the 4 outputs of a {@code DetectionPostProcess} op, i.e:
+ *       <ul>
+ *         <li>Location tensor ({@code kTfLiteFloat32}):
+ *             <ul>
+ *               <li>tensor of size {@code [1 x num_results x 4]}, the inner array representing
+ *                   bounding boxes in the form [top, left, right, bottom].
+ *               <li>{@code BoundingBoxProperties} are required to be attached to the metadata and
+ *                   must specify {@code type=BOUNDARIES} and {@code coordinate_type=RATIO}.
+ *             </ul>
+ *         <li>Classes tensor ({@code kTfLiteFloat32}):
+ *             <ul>
+ *               <li>tensor of size {@code [1 x num_results]}, each value representing the integer
+ *                   index of a class.
+ *               <li>if label maps are attached to the metadata as {@code TENSOR_VALUE_LABELS}
+ *                   associated files, they are used to convert the tensor values into labels.
+ *             </ul>
+ *         <li>scores tensor ({@code kTfLiteFloat32}):
+ *             <ul>
+ *               <li>tensor of size {@code [1 x num_results]}, each value representing the score of
+ *                   the detected object.
+ *             </ul>
+ *         <li>Number of detection tensor ({@code kTfLiteFloat32}):
+ *             <ul>
+ *               <li>integer num_results as a tensor of size {@code [1]}.
+ *             </ul>
+ *       </ul>
+ * </ul>
+ *
+ * <p>An example of such model can be found on <a
+ * href="https://tfhub.dev/google/lite-model/object_detection/mobile_object_localizer_v1/1/metadata/1">TensorFlow
+ * Hub.</a>.
+ */
+public final class ObjectDetector extends BaseTaskApi {
+
+  private static final String OBJECT_DETECTOR_NATIVE_LIB = "task_vision_jni";
+  private static final int OPTIONAL_FD_LENGTH = -1;
+  private static final int OPTIONAL_FD_OFFSET = -1;
+
+  /**
+   * Creates an {@link ObjectDetector} instance from the default {@link ObjectDetectorOptions}.
+   *
+   * @param modelPath path to the detection model with metadata in the assets
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ObjectDetector} from the native
+   *     code
+   */
+  public static ObjectDetector createFromFile(Context context, String modelPath)
+      throws IOException {
+    return createFromFileAndOptions(context, modelPath, ObjectDetectorOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ObjectDetector} instance from the default {@link ObjectDetectorOptions}.
+   *
+   * @param modelFile the detection model {@link File} instance
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ObjectDetector} from the native
+   *     code
+   */
+  public static ObjectDetector createFromFile(File modelFile) throws IOException {
+    return createFromFileAndOptions(modelFile, ObjectDetectorOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ObjectDetector} instance with a model buffer and the default {@link
+   * ObjectDetectorOptions}.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the
+   *     classification model
+   * @throws AssertionError if error occurs when creating {@link ObjectDetector} from the native
+   *     code
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static ObjectDetector createFromBuffer(final ByteBuffer modelBuffer) {
+    return createFromBufferAndOptions(modelBuffer, ObjectDetectorOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ObjectDetector} instance from {@link ObjectDetectorOptions}.
+   *
+   * @param modelPath path to the detection model with metadata in the assets
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ObjectDetector} from the native
+   *     code
+   */
+  public static ObjectDetector createFromFileAndOptions(
+      Context context, String modelPath, ObjectDetectorOptions options) throws IOException {
+    return new ObjectDetector(
+        TaskJniUtils.createHandleFromFdAndOptions(
+            context,
+            new FdAndOptionsHandleProvider<ObjectDetectorOptions>() {
+              @Override
+              public long createHandle(
+                  int fileDescriptor,
+                  long fileDescriptorLength,
+                  long fileDescriptorOffset,
+                  ObjectDetectorOptions options) {
+                return initJniWithModelFdAndOptions(
+                    fileDescriptor, fileDescriptorLength, fileDescriptorOffset, options);
+              }
+            },
+            OBJECT_DETECTOR_NATIVE_LIB,
+            modelPath,
+            options));
+  }
+
+  /**
+   * Creates an {@link ObjectDetector} instance from {@link ObjectDetectorOptions}.
+   *
+   * @param modelFile the detection model {@link File} instance
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ObjectDetector} from the native
+   *     code
+   */
+  public static ObjectDetector createFromFileAndOptions(
+      File modelFile, final ObjectDetectorOptions options) throws IOException {
+    try (ParcelFileDescriptor descriptor =
+        ParcelFileDescriptor.open(modelFile, ParcelFileDescriptor.MODE_READ_ONLY)) {
+      return new ObjectDetector(
+          TaskJniUtils.createHandleFromLibrary(
+              new TaskJniUtils.EmptyHandleProvider() {
+                @Override
+                public long createHandle() {
+                  return initJniWithModelFdAndOptions(
+                      descriptor.getFd(),
+                      /*fileDescriptorLength=*/ OPTIONAL_FD_LENGTH,
+                      /*fileDescriptorOffset=*/ OPTIONAL_FD_OFFSET,
+                      options);
+                }
+              },
+              OBJECT_DETECTOR_NATIVE_LIB));
+    }
+  }
+
+  /**
+   * Creates an {@link ObjectDetector} instance with a model buffer and {@link
+   * ObjectDetectorOptions}.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the
+   *     classification model
+   * @throws AssertionError if error occurs when creating {@link ObjectDetector} from the native
+   *     code
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static ObjectDetector createFromBufferAndOptions(
+      final ByteBuffer modelBuffer, final ObjectDetectorOptions options) {
+    if (!(modelBuffer.isDirect() || modelBuffer instanceof MappedByteBuffer)) {
+      throw new IllegalArgumentException(
+          "The model buffer should be either a direct ByteBuffer or a MappedByteBuffer.");
+    }
+    return new ObjectDetector(
+        TaskJniUtils.createHandleFromLibrary(
+            new EmptyHandleProvider() {
+              @Override
+              public long createHandle() {
+                return initJniWithByteBuffer(modelBuffer, options);
+              }
+            },
+            OBJECT_DETECTOR_NATIVE_LIB));
+  }
+
+  /**
+   * Constructor to initialize the JNI with a pointer from C++.
+   *
+   * @param nativeHandle a pointer referencing memory allocated in C++
+   */
+  private ObjectDetector(long nativeHandle) {
+    super(nativeHandle);
+  }
+
+  /** Options for setting up an ObjectDetector. */
+  @UsedByReflection("object_detector_jni.cc")
+  public static class ObjectDetectorOptions {
+    // Not using AutoValue for this class because scoreThreshold cannot have default value
+    // (otherwise, the default value would override the one in the model metadata) and `Optional` is
+    // not an option here, because
+    // 1. java.util.Optional require Java 8 while we need to support Java 7.
+    // 2. The Guava library (com.google.common.base.Optional) is avoided in this project. See the
+    // comments for labelAllowList.
+    private final String displayNamesLocale;
+    private final int maxResults;
+    private final float scoreThreshold;
+    private final boolean isScoreThresholdSet;
+    // As an open source project, we've been trying avoiding depending on common java libraries,
+    // such as Guava, because it may introduce conflicts with clients who also happen to use those
+    // libraries. Therefore, instead of using ImmutableList here, we convert the List into
+    // unmodifiableList in setLabelAllowList() and setLabelDenyList() to make it less
+    // vulnerable.
+    private final List<String> labelAllowList;
+    private final List<String> labelDenyList;
+    private final int numThreads;
+
+    public static Builder builder() {
+      return new Builder();
+    }
+
+    /** A builder that helps to configure an instance of ObjectDetectorOptions. */
+    public static class Builder {
+      private String displayNamesLocale = "en";
+      private int maxResults = -1;
+      private float scoreThreshold;
+      private boolean isScoreThresholdSet = false;
+      private List<String> labelAllowList = new ArrayList<>();
+      private List<String> labelDenyList = new ArrayList<>();
+      private int numThreads = -1;
+
+      private Builder() {}
+
+      /**
+       * Sets the locale to use for display names specified through the TFLite Model Metadata, if
+       * any.
+       *
+       * <p>Defaults to English({@code "en"}). See the <a
+       * href="https://github.com/tensorflow/tflite-support/blob/3ce83f0cfe2c68fecf83e019f2acc354aaba471f/tensorflow_lite_support/metadata/metadata_schema.fbs#L147">TFLite
+       * Metadata schema file.</a> for the accepted pattern of locale.
+       */
+      public Builder setDisplayNamesLocale(String displayNamesLocale) {
+        this.displayNamesLocale = displayNamesLocale;
+        return this;
+      }
+
+      /**
+       * Sets the maximum number of top-scored detection results to return.
+       *
+       * <p>If < 0, all available results will be returned. If 0, an invalid argument error is
+       * returned. Note that models may intrinsically be limited to returning a maximum number of
+       * results N: if the provided value here is above N, only N results will be returned. Defaults
+       * to -1.
+       *
+       * @throws IllegalArgumentException if maxResults is 0.
+       */
+      public Builder setMaxResults(int maxResults) {
+        if (maxResults == 0) {
+          throw new IllegalArgumentException("maxResults cannot be 0.");
+        }
+        this.maxResults = maxResults;
+        return this;
+      }
+
+      /**
+       * Sets the score threshold that overrides the one provided in the model metadata (if any).
+       * Results below this value are rejected.
+       */
+      public Builder setScoreThreshold(float scoreThreshold) {
+        this.scoreThreshold = scoreThreshold;
+        this.isScoreThresholdSet = true;
+        return this;
+      }
+
+      /**
+       * Sets the optional allow list of labels.
+       *
+       * <p>If non-empty, detection results whose label is not in this set will be filtered out.
+       * Duplicate or unknown labels are ignored. Mutually exclusive with {@code labelDenyList}. It
+       * will cause {@link AssertionError} when calling {@link #createFromFileAndOptions}, if both
+       * {@code labelDenyList} and {@code labelAllowList} are set.
+       */
+      public Builder setLabelAllowList(List<String> labelAllowList) {
+        this.labelAllowList = Collections.unmodifiableList(new ArrayList<>(labelAllowList));
+        return this;
+      }
+
+      /**
+       * Sets the optional deny list of labels.
+       *
+       * <p>If non-empty, detection results whose label is in this set will be filtered out.
+       * Duplicate or unknown labels are ignored. Mutually exclusive with {@code labelAllowList}. It
+       * will cause {@link AssertionError} when calling {@link #createFromFileAndOptions}, if both
+       * {@code labelDenyList} and {@code labelAllowList} are set.
+       */
+      public Builder setLabelDenyList(List<String> labelDenyList) {
+        this.labelDenyList = Collections.unmodifiableList(new ArrayList<>(labelDenyList));
+        return this;
+      }
+
+      /**
+       * Sets the number of threads to be used for TFLite ops that support multi-threading when
+       * running inference with CPU. Defaults to -1.
+       *
+       * <p>numThreads should be greater than 0 or equal to -1. Setting numThreads to -1 has the
+       * effect to let TFLite runtime set the value.
+       */
+      public Builder setNumThreads(int numThreads) {
+        this.numThreads = numThreads;
+        return this;
+      }
+
+      public ObjectDetectorOptions build() {
+        return new ObjectDetectorOptions(this);
+      }
+    }
+
+    @UsedByReflection("object_detector_jni.cc")
+    public String getDisplayNamesLocale() {
+      return displayNamesLocale;
+    }
+
+    @UsedByReflection("object_detector_jni.cc")
+    public int getMaxResults() {
+      return maxResults;
+    }
+
+    @UsedByReflection("object_detector_jni.cc")
+    public float getScoreThreshold() {
+      return scoreThreshold;
+    }
+
+    @UsedByReflection("object_detector_jni.cc")
+    public boolean getIsScoreThresholdSet() {
+      return isScoreThresholdSet;
+    }
+
+    @UsedByReflection("object_detector_jni.cc")
+    public List<String> getLabelAllowList() {
+      return new ArrayList<>(labelAllowList);
+    }
+
+    @UsedByReflection("object_detector_jni.cc")
+    public List<String> getLabelDenyList() {
+      return new ArrayList<>(labelDenyList);
+    }
+
+    @UsedByReflection("object_detector_jni.cc")
+    public int getNumThreads() {
+      return numThreads;
+    }
+
+    private ObjectDetectorOptions(Builder builder) {
+      displayNamesLocale = builder.displayNamesLocale;
+      maxResults = builder.maxResults;
+      scoreThreshold = builder.scoreThreshold;
+      isScoreThresholdSet = builder.isScoreThresholdSet;
+      labelAllowList = builder.labelAllowList;
+      labelDenyList = builder.labelDenyList;
+      numThreads = builder.numThreads;
+    }
+  }
+
+  /**
+   * Performs actual detection on the provided image.
+   *
+   * @param image a {@link TensorImage} object that represents a RGB image
+   * @throws AssertionError if error occurs when processing the image from the native code
+   */
+  public List<Detection> detect(TensorImage image) {
+    return detect(image, ImageProcessingOptions.builder().build());
+  }
+
+  /**
+   * Performs actual detection on the provided image.
+   *
+   * @param image a {@link TensorImage} object that represents a RGB image
+   * @param options {@link ObjectDetector} only supports image rotation (through {@link
+   *     ImageProcessingOptions#Builder#setOrientation}) currently. The orientation of an image
+   *     defaults to {@link ImageProcessingOptions#Orientation#TOP_LEFT}.
+   * @throws AssertionError if error occurs when processing the image from the native code
+   */
+  public List<Detection> detect(TensorImage image, ImageProcessingOptions options) {
+    checkNotClosed();
+
+    // object_detector_jni.cc expects an uint8 image. Convert image of other types into uint8.
+    TensorImage imageUint8 =
+        image.getDataType() == DataType.UINT8
+            ? image
+            : TensorImage.createFrom(image, DataType.UINT8);
+    return detectNative(
+        getNativeHandle(),
+        imageUint8.getBuffer(),
+        imageUint8.getWidth(),
+        imageUint8.getHeight(),
+        options.getOrientation().getValue());
+  }
+
+  private static native long initJniWithModelFdAndOptions(
+      int fileDescriptor,
+      long fileDescriptorLength,
+      long fileDescriptorOffset,
+      ObjectDetectorOptions options);
+
+  private static native long initJniWithByteBuffer(
+      ByteBuffer modelBuffer, ObjectDetectorOptions options);
+
+  private static native List<Detection> detectNative(
+      long nativeHandle, ByteBuffer image, int width, int height, int orientation);
+
+  @Override
+  protected void deinit(long nativeHandle) {
+    deinitJni(nativeHandle);
+  }
+
+  /**
+   * Native implementation to release memory pointed by the pointer.
+   *
+   * @param nativeHandle pointer to memory allocated
+   */
+  private native void deinitJni(long nativeHandle);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/AndroidManifest.xml b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/AndroidManifest.xml
new file mode 100644
index 0000000..991d481
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/AndroidManifest.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="org.tensorflow.lite.task.vision.segmenter">
+  <uses-sdk android:minSdkVersion="19" android:targetSdkVersion="29"/>
+</manifest>
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/BUILD b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/BUILD
new file mode 100644
index 0000000..506d5bf
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/BUILD
@@ -0,0 +1,41 @@
+load("@org_tensorflow//tensorflow/lite/java:aar_with_jni.bzl", "aar_with_jni")
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "AndroidManifest.xml",
+])
+
+filegroup(
+    name = "image_segmenter_src",
+    srcs = glob(["**/*.java"]),
+)
+
+android_library(
+    name = "image_segmenter",
+    srcs = glob(["*.java"]),
+    # TODO(b/163039980): Use JAVACOPTS in TF. "-Xep:RemoveUnusedImports:ERROR" wierdly break the build.
+    javacopts = ["-source 7 -target 7"],
+    manifest = "AndroidManifest.xml",
+    deps = [
+        "//tensorflow_lite_support/java:tensorflowlite_support_java",
+        "//tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base_task_api",
+        "//tensorflow_lite_support/java/src/native/task/vision/segmenter:image_segmenter_native",
+        "@com_google_auto_value",
+        "@maven//:androidx_annotation_annotation",
+        "@org_tensorflow//tensorflow/lite/java:tensorflowlite_java",
+    ],
+)
+
+# AAR target for OSS release.
+#
+# bazel build -c opt --config=monolithic --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+#   tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter:image-segmenter
+aar_with_jni(
+    name = "image-segmenter",
+    android_library = ":image_segmenter",
+)
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/ColoredLabel.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/ColoredLabel.java
new file mode 100644
index 0000000..09416d0
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/ColoredLabel.java
@@ -0,0 +1,88 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.segmenter;
+
+import android.graphics.Color;
+import android.os.Build;
+import androidx.annotation.RequiresApi;
+import com.google.auto.value.AutoValue;
+import org.tensorflow.lite.annotations.UsedByReflection;
+
+/** Represents a label associated with a color for display purposes. */
+@AutoValue
+@UsedByReflection("image_segmentation_jni.cc")
+public abstract class ColoredLabel {
+
+  /**
+   * Creates a {@link ColoredLabel} object with an ARGB color int.
+   *
+   * @param label the label string, as provided in the label map packed in the TFLite Model
+   *     Metadata.
+   * @param displayName the display name of label, as configured through {@link
+   *     ImageSegmenter#ImageSegmenterOptions#Builder#setDisplayNamesLocale}
+   * @param argb the color components for the label in ARGB. See <a
+   *     href="https://developer.android.com/reference/android/graphics/Color#color-ints">Android
+   *     Color ints.</a> for more details.
+   */
+  @UsedByReflection("image_segmentation_jni.cc")
+  public static ColoredLabel create(String label, String displayName, int argb) {
+    return new AutoValue_ColoredLabel(label, displayName, argb);
+  }
+
+  /**
+   * Creates a {@link ColoredLabel} object with a {@link Color} instance.
+   *
+   * @param label the label string, as provided in the label map packed in the TFLite Model
+   *     Metadata.
+   * @param displayName the display name of label, as configured through {@link
+   *     ImageSegmenter#ImageSegmenterOptions#Builder#setDisplayNamesLocale}
+   * @param color the color components for the label. The Color instatnce is supported on Android
+   *     API level 26 and above. For API level lower than 26, use {@link #create(String, String,
+   *     int)}. See <a
+   *     href="https://developer.android.com/reference/android/graphics/Color#color-instances">Android
+   *     Color instances.</a> for more details.
+   */
+  @RequiresApi(Build.VERSION_CODES.O)
+  public static ColoredLabel create(String label, String displayName, Color color) {
+    return new AutoValue_ColoredLabel(label, displayName, color.toArgb());
+  }
+
+  public abstract String getlabel();
+
+  public abstract String getDisplayName();
+
+  /**
+   * Gets the ARGB int that represents the color.
+   *
+   * <p>See <a
+   * href="https://developer.android.com/reference/android/graphics/Color#color-ints">Android Color
+   * ints.</a> for more details.
+   */
+  public abstract int getArgb();
+
+  /**
+   * Gets the {@link Color} instance of the underlying color.
+   *
+   * <p>The Color instatnce is supported on Android API level 26 and above. For API level lower than
+   * 26, use {@link #getArgb()}. See <a
+   * href="https://developer.android.com/reference/android/graphics/Color#color-instances">Android
+   * Color instances.</a> for more details.
+   */
+  @RequiresApi(Build.VERSION_CODES.O)
+  public Color getColor() {
+    return Color.valueOf(getArgb());
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/ImageSegmenter.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/ImageSegmenter.java
new file mode 100644
index 0000000..bd90790
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/ImageSegmenter.java
@@ -0,0 +1,377 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.segmenter;
+
+import android.content.Context;
+import android.content.res.AssetFileDescriptor;
+import android.os.ParcelFileDescriptor;
+import com.google.auto.value.AutoValue;
+import java.io.File;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.ByteOrder;
+import java.nio.MappedByteBuffer;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.List;
+import org.tensorflow.lite.DataType;
+import org.tensorflow.lite.support.image.TensorImage;
+import org.tensorflow.lite.task.core.BaseTaskApi;
+import org.tensorflow.lite.task.core.TaskJniUtils;
+import org.tensorflow.lite.task.core.TaskJniUtils.EmptyHandleProvider;
+import org.tensorflow.lite.task.core.vision.ImageProcessingOptions;
+
+/**
+ * Performs segmentation on images.
+ *
+ * <p>The API expects a TFLite model with <a
+ * href="https://www.tensorflow.org/lite/convert/metadata">TFLite Model Metadata.</a>.
+ *
+ * <p>The API supports models with one image input tensor and one output tensor. To be more
+ * specific, here are the requirements.
+ *
+ * <ul>
+ *   <li>Input image tensor ({@code kTfLiteUInt8}/{@code kTfLiteFloat32})
+ *       <ul>
+ *         <li>image input of size {@code [batch x height x width x channels]}.
+ *         <li>batch inference is not supported ({@code batch} is required to be 1).
+ *         <li>only RGB inputs are supported ({@code channels} is required to be 3).
+ *         <li>if type is {@code kTfLiteFloat32}, NormalizationOptions are required to be attached
+ *             to the metadata for input normalization.
+ *       </ul>
+ *   <li>Output image tensor ({@code kTfLiteUInt8}/{@code kTfLiteFloat32})
+ *       <ul>
+ *         <li>tensor of size {@code [batch x mask_height x mask_width x num_classes]}, where {@code
+ *             batch} is required to be 1, {@code mask_width} and {@code mask_height} are the
+ *             dimensions of the segmentation masks produced by the model, and {@code num_classes}
+ *             is the number of classes supported by the model.
+ *         <li>optional (but recommended) label map(s) can be attached as AssociatedFile-s with type
+ *             TENSOR_AXIS_LABELS, containing one label per line. The first such AssociatedFile (if
+ *             any) is used to fill the class name, i.e. {@link ColoredLabel#getClassName} of the
+ *             results. The display name, i.e. {@link ColoredLabel#getDisplayName}, is filled from
+ *             the AssociatedFile (if any) whose locale matches the `display_names_locale` field of
+ *             the `ImageSegmenterOptions` used at creation time ("en" by default, i.e. English). If
+ *             none of these are available, only the `index` field of the results will be filled.
+ *       </ul>
+ * </ul>
+ *
+ * <p>An example of such model can be found on <a
+ * href="https://tfhub.dev/tensorflow/lite-model/deeplabv3/1/metadata/1">TensorFlow Hub.</a>.
+ */
+public final class ImageSegmenter extends BaseTaskApi {
+
+  private static final String IMAGE_SEGMENTER_NATIVE_LIB = "task_vision_jni";
+  private static final int OPTIONAL_FD_LENGTH = -1;
+  private static final int OPTIONAL_FD_OFFSET = -1;
+
+  private final OutputType outputType;
+
+  /**
+   * Creates an {@link ImageSegmenter} instance from the default {@link ImageSegmenterOptions}.
+   *
+   * @param modelPath path of the segmentation model with metadata in the assets
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageSegmenter} from the native
+   *     code
+   */
+  public static ImageSegmenter createFromFile(Context context, String modelPath)
+      throws IOException {
+    return createFromFileAndOptions(context, modelPath, ImageSegmenterOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ImageSegmenter} instance from the default {@link ImageSegmenterOptions}.
+   *
+   * @param modelFile the segmentation model {@link File} instance
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageSegmenter} from the native
+   *     code
+   */
+  public static ImageSegmenter createFromFile(File modelFile) throws IOException {
+    return createFromFileAndOptions(modelFile, ImageSegmenterOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ImageSegmenter} instance with a model buffer and the default {@link
+   * ImageSegmenterOptions}.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the
+   *     classification model
+   * @throws AssertionError if error occurs when creating {@link ImageSegmenter} from the native
+   *     code
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static ImageSegmenter createFromBuffer(final ByteBuffer modelBuffer) {
+    return createFromBufferAndOptions(modelBuffer, ImageSegmenterOptions.builder().build());
+  }
+
+  /**
+   * Creates an {@link ImageSegmenter} instance from {@link ImageSegmenterOptions}.
+   *
+   * @param modelPath path of the segmentation model with metadata in the assets
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageSegmenter} from the native
+   *     code
+   */
+  public static ImageSegmenter createFromFileAndOptions(
+      Context context, String modelPath, final ImageSegmenterOptions options) throws IOException {
+    try (AssetFileDescriptor assetFileDescriptor = context.getAssets().openFd(modelPath)) {
+      return createFromModelFdAndOptions(
+          /*fileDescriptor=*/ assetFileDescriptor.getParcelFileDescriptor().getFd(),
+          /*fileDescriptorLength=*/ assetFileDescriptor.getLength(),
+          /*fileDescriptorOffset=*/ assetFileDescriptor.getStartOffset(),
+          options);
+    }
+  }
+
+  /**
+   * Creates an {@link ImageSegmenter} instance from {@link ImageSegmenterOptions}.
+   *
+   * @param modelFile the segmentation model {@link File} instance
+   * @throws IOException if an I/O error occurs when loading the tflite model
+   * @throws AssertionError if error occurs when creating {@link ImageSegmenter} from the native
+   *     code
+   */
+  public static ImageSegmenter createFromFileAndOptions(
+      File modelFile, final ImageSegmenterOptions options) throws IOException {
+    try (ParcelFileDescriptor descriptor =
+        ParcelFileDescriptor.open(modelFile, ParcelFileDescriptor.MODE_READ_ONLY)) {
+      return createFromModelFdAndOptions(
+          /*fileDescriptor=*/ descriptor.getFd(),
+          /*fileDescriptorLength=*/ OPTIONAL_FD_LENGTH,
+          /*fileDescriptorOffset=*/ OPTIONAL_FD_OFFSET,
+          options);
+    }
+  }
+
+  /**
+   * Creates an {@link ImageSegmenter} instance with a model buffer and {@link
+   * ImageSegmenterOptions}.
+   *
+   * @param modelBuffer a direct {@link ByteBuffer} or a {@link MappedByteBuffer} of the
+   *     classification model
+   * @throws AssertionError if error occurs when creating {@link ImageSegmenter} from the native
+   *     code
+   * @throws IllegalArgumentException if the model buffer is not a direct {@link ByteBuffer} or a
+   *     {@link MappedByteBuffer}
+   */
+  public static ImageSegmenter createFromBufferAndOptions(
+      final ByteBuffer modelBuffer, final ImageSegmenterOptions options) {
+    if (!(modelBuffer.isDirect() || modelBuffer instanceof MappedByteBuffer)) {
+      throw new IllegalArgumentException(
+          "The model buffer should be either a direct ByteBuffer or a MappedByteBuffer.");
+    }
+    return new ImageSegmenter(
+        TaskJniUtils.createHandleFromLibrary(
+            new EmptyHandleProvider() {
+              @Override
+              public long createHandle() {
+                return initJniWithByteBuffer(
+                    modelBuffer,
+                    options.getDisplayNamesLocale(),
+                    options.getOutputType().getValue(),
+                    options.getNumThreads());
+              }
+            },
+            IMAGE_SEGMENTER_NATIVE_LIB),
+        options.getOutputType());
+  }
+
+  /**
+   * Constructor to initialize the JNI with a pointer from C++.
+   *
+   * @param nativeHandle a pointer referencing memory allocated in C++
+   */
+  private ImageSegmenter(long nativeHandle, OutputType outputType) {
+    super(nativeHandle);
+    this.outputType = outputType;
+  }
+
+  /** Options for setting up an {@link ImageSegmenter}. */
+  @AutoValue
+  public abstract static class ImageSegmenterOptions {
+    private static final String DEFAULT_DISPLAY_NAME_LOCALE = "en";
+    private static final OutputType DEFAULT_OUTPUT_TYPE = OutputType.CATEGORY_MASK;
+    private static final int NUM_THREADS = -1;
+
+    public abstract String getDisplayNamesLocale();
+
+    public abstract OutputType getOutputType();
+
+    public abstract int getNumThreads();
+
+    public static Builder builder() {
+      return new AutoValue_ImageSegmenter_ImageSegmenterOptions.Builder()
+          .setDisplayNamesLocale(DEFAULT_DISPLAY_NAME_LOCALE)
+          .setOutputType(DEFAULT_OUTPUT_TYPE)
+          .setNumThreads(NUM_THREADS);
+    }
+
+    /** Builder for {@link ImageSegmenterOptions}. */
+    @AutoValue.Builder
+    public abstract static class Builder {
+
+      /**
+       * Sets the locale to use for display names specified through the TFLite Model Metadata, if
+       * any.
+       *
+       * <p>Defaults to English({@code "en"}). See the <a
+       * href="https://github.com/tensorflow/tflite-support/blob/3ce83f0cfe2c68fecf83e019f2acc354aaba471f/tensorflow_lite_support/metadata/metadata_schema.fbs#L147">TFLite
+       * Metadata schema file.</a> for the accepted pattern of locale.
+       */
+      public abstract Builder setDisplayNamesLocale(String displayNamesLocale);
+
+      public abstract Builder setOutputType(OutputType outputType);
+
+      /**
+       * Sets the number of threads to be used for TFLite ops that support multi-threading when
+       * running inference with CPU. Defaults to -1.
+       *
+       * <p>numThreads should be greater than 0 or equal to -1. Setting numThreads to -1 has the
+       * effect to let TFLite runtime set the value.
+       */
+      public abstract Builder setNumThreads(int numThreads);
+
+      public abstract ImageSegmenterOptions build();
+    }
+  }
+
+  /**
+   * Performs actual segmentation on the provided image.
+   *
+   * @param image a {@link TensorImage} object that represents an RGB image
+   * @return results of performing image segmentation. Note that at the time, a single {@link
+   *     Segmentation} element is expected to be returned. The result is stored in a {@link List}
+   *     for later extension to e.g. instance segmentation models, which may return one segmentation
+   *     per object.
+   * @throws AssertionError if error occurs when segmenting the image from the native code
+   */
+  public List<Segmentation> segment(TensorImage image) {
+    return segment(image, ImageProcessingOptions.builder().build());
+  }
+
+  /**
+   * Performs actual segmentation on the provided image with {@link ImageProcessingOptions}.
+   *
+   * @param image a {@link TensorImage} object that represents an RGB image
+   * @param options {@link ImageSegmenter} only supports image rotation (through {@link
+   *     ImageProcessingOptions#Builder#setOrientation}) currently. The orientation of an image
+   *     defaults to {@link ImageProcessingOptions#Orientation#TOP_LEFT}.
+   * @return results of performing image segmentation. Note that at the time, a single {@link
+   *     Segmentation} element is expected to be returned. The result is stored in a {@link List}
+   *     for later extension to e.g. instance segmentation models, which may return one segmentation
+   *     per object.
+   * @throws AssertionError if error occurs when segmenting the image from the native code
+   */
+  public List<Segmentation> segment(TensorImage image, ImageProcessingOptions options) {
+    checkNotClosed();
+
+    // image_segmenter_jni.cc expects an uint8 image. Convert image of other types into uint8.
+    TensorImage imageUint8 =
+        image.getDataType() == DataType.UINT8
+            ? image
+            : TensorImage.createFrom(image, DataType.UINT8);
+    List<byte[]> maskByteArrays = new ArrayList<>();
+    List<ColoredLabel> coloredLabels = new ArrayList<>();
+    int[] maskShape = new int[2];
+    segmentNative(
+        getNativeHandle(),
+        imageUint8.getBuffer(),
+        imageUint8.getWidth(),
+        imageUint8.getHeight(),
+        maskByteArrays,
+        maskShape,
+        coloredLabels,
+        options.getOrientation().getValue());
+
+    List<ByteBuffer> maskByteBuffers = new ArrayList<>();
+    for (byte[] bytes : maskByteArrays) {
+      ByteBuffer byteBuffer = ByteBuffer.wrap(bytes);
+      // Change the byte order to little_endian, since the buffers were generated in jni.
+      byteBuffer.order(ByteOrder.LITTLE_ENDIAN);
+      maskByteBuffers.add(byteBuffer);
+    }
+
+    return Arrays.asList(
+        Segmentation.create(
+            outputType,
+            outputType.createMasksFromBuffer(maskByteBuffers, maskShape),
+            coloredLabels));
+  }
+
+  private static ImageSegmenter createFromModelFdAndOptions(
+      final int fileDescriptor,
+      final long fileDescriptorLength,
+      final long fileDescriptorOffset,
+      final ImageSegmenterOptions options) {
+    long nativeHandle =
+        TaskJniUtils.createHandleFromLibrary(
+            new EmptyHandleProvider() {
+              @Override
+              public long createHandle() {
+                return initJniWithModelFdAndOptions(
+                    fileDescriptor,
+                    fileDescriptorLength,
+                    fileDescriptorOffset,
+                    options.getDisplayNamesLocale(),
+                    options.getOutputType().getValue(),
+                    options.getNumThreads());
+              }
+            },
+            IMAGE_SEGMENTER_NATIVE_LIB);
+    return new ImageSegmenter(nativeHandle, options.getOutputType());
+  }
+
+  private static native long initJniWithModelFdAndOptions(
+      int fileDescriptor,
+      long fileDescriptorLength,
+      long fileDescriptorOffset,
+      String displayNamesLocale,
+      int outputType,
+      int numThreads);
+
+  private static native long initJniWithByteBuffer(
+      ByteBuffer modelBuffer, String displayNamesLocale, int outputType, int numThreads);
+
+  /**
+   * The native method to segment the image.
+   *
+   * <p>{@code maskBuffers}, {@code maskShape}, {@code coloredLabels} will be updated in the native
+   * layer.
+   */
+  private static native void segmentNative(
+      long nativeHandle,
+      ByteBuffer image,
+      int width,
+      int height,
+      List<byte[]> maskByteArrays,
+      int[] maskShape,
+      List<ColoredLabel> coloredLabels,
+      int orientation);
+
+  @Override
+  protected void deinit(long nativeHandle) {
+    deinitJni(nativeHandle);
+  }
+
+  /**
+   * Native implementation to release memory pointed by the pointer.
+   *
+   * @param nativeHandle pointer to memory allocated
+   */
+  private native void deinitJni(long nativeHandle);
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/OutputType.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/OutputType.java
new file mode 100644
index 0000000..03d82c6
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/OutputType.java
@@ -0,0 +1,145 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.segmenter;
+
+import static org.tensorflow.lite.DataType.FLOAT32;
+import static org.tensorflow.lite.DataType.UINT8;
+import static org.tensorflow.lite.support.common.SupportPreconditions.checkArgument;
+import static org.tensorflow.lite.support.image.ColorSpaceType.GRAYSCALE;
+
+import java.nio.ByteBuffer;
+import java.util.ArrayList;
+import java.util.List;
+import org.tensorflow.lite.support.image.TensorImage;
+import org.tensorflow.lite.support.tensorbuffer.TensorBuffer;
+
+/**
+ * Output mask type. This allows specifying the type of post-processing to perform on the raw model
+ * results.
+ */
+public enum OutputType {
+
+  /**
+   * Gives a single output mask where each pixel represents the class which the pixel in the
+   * original image was predicted to belong to.
+   */
+  CATEGORY_MASK(0) {
+    /**
+     * {@inheritDoc}
+     *
+     * @throws IllegalArgumentException if more than one {@link TensorImage} are provided, or if the
+     *     color space of the {@link TensorImage} is not {@link ColorSpaceType#GRAYSCALE}
+     */
+    @Override
+    void assertMasksMatchColoredLabels(List<TensorImage> masks, List<ColoredLabel> coloredLabels) {
+      checkArgument(
+          masks.size() == 1,
+          "CATRGORY_MASK only allows one TensorImage in the list, providing " + masks.size());
+
+      TensorImage mask = masks.get(0);
+      checkArgument(
+          mask.getColorSpaceType() == GRAYSCALE,
+          "CATRGORY_MASK only supports masks of ColorSpaceType, GRAYSCALE, providing "
+              + mask.getColorSpaceType());
+    }
+
+    /**
+     * {@inheritDoc}
+     *
+     * @throws IllegalArgumentException if more than one {@link ByteBuffer} are provided in the list
+     */
+    @Override
+    List<TensorImage> createMasksFromBuffer(List<ByteBuffer> buffers, int[] maskShape) {
+      checkArgument(
+          buffers.size() == 1,
+          "CATRGORY_MASK only allows one mask in the buffer list, providing " + buffers.size());
+
+      List<TensorImage> masks = new ArrayList<>();
+      TensorBuffer tensorBuffer = TensorBuffer.createDynamic(UINT8);
+      tensorBuffer.loadBuffer(buffers.get(0), maskShape);
+      TensorImage tensorImage = new TensorImage(UINT8);
+      tensorImage.load(tensorBuffer, GRAYSCALE);
+      masks.add(tensorImage);
+
+      return masks;
+    }
+  },
+
+  /**
+   * Gives a list of output masks where, for each mask, each pixel represents the prediction
+   * confidence, usually in the [0, 1] range.
+   */
+  CONFIDENCE_MASK(1) {
+    /**
+     * {@inheritDoc}
+     *
+     * @throws IllegalArgumentException if more the size of the masks list does not match the size
+     *     of the coloredlabels list, or if the color space type of the any mask is not {@link
+     *     ColorSpaceType#GRAYSCALE}
+     */
+    @Override
+    void assertMasksMatchColoredLabels(List<TensorImage> masks, List<ColoredLabel> coloredLabels) {
+      checkArgument(
+          masks.size() == coloredLabels.size(),
+          String.format(
+              "When using CONFIDENCE_MASK, the number of masks (%d) should match the number of"
+                  + " coloredLabels (%d).",
+              masks.size(), coloredLabels.size()));
+
+      for (TensorImage mask : masks) {
+        checkArgument(
+            mask.getColorSpaceType() == GRAYSCALE,
+            "CONFIDENCE_MASK only supports masks of ColorSpaceType, GRAYSCALE, providing "
+                + mask.getColorSpaceType());
+      }
+    }
+
+    @Override
+    List<TensorImage> createMasksFromBuffer(List<ByteBuffer> buffers, int[] maskShape) {
+      List<TensorImage> masks = new ArrayList<>();
+      for (ByteBuffer buffer : buffers) {
+        TensorBuffer tensorBuffer = TensorBuffer.createDynamic(FLOAT32);
+        tensorBuffer.loadBuffer(buffer, maskShape);
+        TensorImage tensorImage = new TensorImage(FLOAT32);
+        tensorImage.load(tensorBuffer, GRAYSCALE);
+        masks.add(tensorImage);
+      }
+      return masks;
+    }
+  };
+
+  public int getValue() {
+    return value;
+  }
+
+  /**
+   * Verifies that the given list of masks matches the list of colored labels.
+   *
+   * @throws IllegalArgumentException if {@code masks} and {@code coloredLabels} do not match the
+   *     output type
+   */
+  abstract void assertMasksMatchColoredLabels(
+      List<TensorImage> masks, List<ColoredLabel> coloredLabels);
+
+  /** Creates the masks in {@link TensorImage} based on the data in {@link ByteBuffer}. */
+  abstract List<TensorImage> createMasksFromBuffer(List<ByteBuffer> buffers, int[] maskShape);
+
+  private final int value;
+
+  private OutputType(int value) {
+    this.value = value;
+  }
+}
diff --git a/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/Segmentation.java b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/Segmentation.java
new file mode 100644
index 0000000..018482c
--- /dev/null
+++ b/tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision/segmenter/Segmentation.java
@@ -0,0 +1,82 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+package org.tensorflow.lite.task.vision.segmenter;
+
+import com.google.auto.value.AutoValue;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+import org.tensorflow.lite.support.image.TensorImage;
+
+/** Represents the segmentation result of an {@link ImageSegmenter}. */
+@AutoValue
+public abstract class Segmentation {
+
+  /**
+   * Creates a {@link Segmentation} object.
+   *
+   * <p>{@link Segmentation} provides two types of outputs as indicated through {@link OutputType}:
+   *
+   * <p>{@link OutputType#CATEGORY_MASK}: the result contains a single category mask, which is a
+   * grayscale {@link TensorImage} with shape (height, width), in row major order. The value of each
+   * pixel in this mask represents the class to which the pixel in the mask belongs. The pixel
+   * values are in 1:1 corresponding with the colored labels, i.e. a pixel with value {@code i} is
+   * associated with {@code coloredLabels.get(i)}.
+   *
+   * <p>{@link OutputType#CONFIDENCE_MASK}: the result contains a list of confidence masks, which
+   * are in 1:1 correspondance with the colored labels, i.e. {@link masks.get(i)} is associated with
+   * {@code coloredLabels.get(i)}. Each confidence mask is a grayscale {@link TensorImage} with
+   * shape (height, width), in row major order. The value of each pixel in these masks represents
+   * the confidence score for this particular class.
+   *
+   * <p>IMPORTANT: segmentation masks are not direcly suited for display, in particular:<br>
+   * \* they are relative to the unrotated input frame, i.e. *not* taking into account the {@code
+   * Orientation} flag of the input FrameBuffer, <br>
+   * \* their dimensions are intrinsic to the model, i.e. *not* dependent on the input FrameBuffer
+   * dimensions.
+   *
+   * <p>Example of such post-processing, assuming: <br>
+   * \* an input FrameBuffer with width=640, height=480, orientation=kLeftBottom (i.e. the image
+   * will be rotated 90° clockwise during preprocessing to make it "upright"), <br>
+   * \* a model outputting masks of size 224x224. <br>
+   * In order to be directly displayable on top of the input image assumed to be displayed *with*
+   * the {@code Orientation} flag taken into account (according to the <a
+   * href="http://jpegclub.org/exif_orientation.html">EXIF specification</a>), the masks need to be:
+   * re-scaled to 640 x 480, then rotated 90° clockwise.
+   *
+   * @throws IllegalArgumentException if {@code masks} and {@code coloredLabels} do not match the
+   *     {@code outputType}
+   */
+  static Segmentation create(
+      OutputType outputType, List<TensorImage> masks, List<ColoredLabel> coloredLabels) {
+    outputType.assertMasksMatchColoredLabels(masks, coloredLabels);
+
+    return new AutoValue_Segmentation(
+        outputType,
+        Collections.unmodifiableList(new ArrayList<TensorImage>(masks)),
+        Collections.unmodifiableList(new ArrayList<ColoredLabel>(coloredLabels)));
+  }
+
+  public abstract OutputType getOutputType();
+
+  // As an open source project, we've been trying avoiding depending on common java libraries,
+  // such as Guava, because it may introduce conflicts with clients who also happen to use those
+  // libraries. Therefore, instead of using ImmutableList here, we convert the List into
+  // unmodifiableList in create() to make it less vulnerable.
+  public abstract List<TensorImage> getMasks();
+
+  public abstract List<ColoredLabel> getColoredLabels();
+}
diff --git a/tensorflow_lite_support/java/src/native/task/core/BUILD b/tensorflow_lite_support/java/src/native/task/core/BUILD
new file mode 100644
index 0000000..d4dd7ab
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/core/BUILD
@@ -0,0 +1,16 @@
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+# Default provider for BuiltInOpResover. Create your own target, overwrite the
+# function to provide a MutableOpResolver for customized OPs and/or a subset of
+# builtin OPs.
+cc_library(
+    name = "builtin_op_resolver",
+    srcs = ["builtin_op_resolver.cc"],
+    deps = [
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/core/builtin_op_resolver.cc b/tensorflow_lite_support/java/src/native/task/core/builtin_op_resolver.cc
new file mode 100644
index 0000000..050f49f
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/core/builtin_op_resolver.cc
@@ -0,0 +1,27 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow/lite/kernels/register.h"
+
+namespace tflite {
+namespace task {
+// Default provider for OpResolver, provides BuiltinOpResolver.
+std::unique_ptr<OpResolver> CreateOpResolver() {  // NOLINT
+  return std::unique_ptr<tflite::ops::builtin::BuiltinOpResolver>(
+      new tflite::ops::builtin::BuiltinOpResolver);
+}
+
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/java/src/native/task/text/BUILD b/tensorflow_lite_support/java/src/native/task/text/BUILD
new file mode 100644
index 0000000..a27aba5
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/BUILD
@@ -0,0 +1,34 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "task_text_native",
+    srcs = [
+        ":libtask_text_jni.so",
+    ],
+)
+
+tflite_jni_binary(
+    name = "libtask_text_jni.so",
+    srcs = [
+        "//tensorflow_lite_support/java/src/native/task/text/nlclassifier:nl_classifier_jni.cc",
+        "//tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier:bert_nl_classifier_jni.cc",
+        "//tensorflow_lite_support/java/src/native/task/text/qa:bert_question_answerer_jni.cc",
+    ],
+    linkscript = "//tensorflow_lite_support/java:default_version_script.lds",
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:bert_nl_classifier",
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier",
+        "//tensorflow_lite_support/cc/task/text/qa:bert_question_answerer",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "//tensorflow_lite_support/java/src/native/task/core:builtin_op_resolver",
+        "//tensorflow_lite_support/java/src/native/task/text/nlclassifier:nl_classifier_jni_utils",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/text/nlclassifier/BUILD b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/BUILD
new file mode 100644
index 0000000..88f3e9b
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/BUILD
@@ -0,0 +1,63 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "nl_classifier_jni.cc",
+])
+
+# Default native target for nl_classifier to provide BuiltInOpResolver.
+cc_library(
+    name = "nl_classifier_native",
+    srcs = [
+        ":libtask_text_jni.so",
+    ],
+)
+
+# Note: "libtask_text_jni" is hardcoded in Java to look up the .so, therefore
+# the name should remain the same when creating customized version of
+# nl_classifier_native
+tflite_jni_binary(
+    name = "libtask_text_jni.so",
+    linkscript = "//tensorflow_lite_support/java:default_version_script.lds",
+    deps = [
+        ":native_without_resolver",
+        "//tensorflow_lite_support/java/src/native/task/core:builtin_op_resolver",
+    ],
+)
+
+# Shared native logic for NLClassifier. Combine this target and customized
+# version of op_resolver to build customized nl_classifier_native target.
+cc_library(
+    name = "native_without_resolver",
+    srcs = [
+        "nl_classifier_jni.cc",
+    ],
+    deps = [
+        ":nl_classifier_jni_utils",
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "@org_tensorflow//tensorflow/lite:framework",
+        "@org_tensorflow//tensorflow/lite/kernels:kernel_util",
+    ],
+    alwayslink = 1,
+)
+
+cc_library(
+    name = "nl_classifier_jni_utils",
+    srcs = [
+        "nl_classifier_jni_utils.cc",
+    ],
+    hdrs = [
+        "nl_classifier_jni_utils.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:nl_classifier",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier/BUILD b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier/BUILD
new file mode 100644
index 0000000..49f3f4e
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier/BUILD
@@ -0,0 +1,31 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "bert_nl_classifier_jni.cc",
+])
+
+cc_library(
+    name = "bert_nl_classifier_native",
+    srcs = [
+        ":libtask_text_jni.so",
+    ],
+)
+
+tflite_jni_binary(
+    name = "libtask_text_jni.so",
+    srcs = [
+        "bert_nl_classifier_jni.cc",
+    ],
+    linkscript = "//tensorflow_lite_support/java:default_version_script.lds",
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/nlclassifier:bert_nl_classifier",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "//tensorflow_lite_support/java/src/native/task/text/nlclassifier:nl_classifier_jni_utils",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier/bert_nl_classifier_jni.cc b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier/bert_nl_classifier_jni.cc
new file mode 100644
index 0000000..1edb350
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/bert_nlclassifier/bert_nl_classifier_jni.cc
@@ -0,0 +1,74 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/bert_nl_classifier.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+#include "tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.h"
+
+namespace {
+
+using ::tflite::support::utils::GetMappedFileBuffer;
+using ::tflite::support::utils::kAssertionError;
+using ::tflite::support::utils::kInvalidPointer;
+using ::tflite::support::utils::ThrowException;
+using ::tflite::task::text::nlclassifier::BertNLClassifier;
+using ::tflite::task::text::nlclassifier::RunClassifier;
+
+extern "C" JNIEXPORT void JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_BertNLClassifier_deinitJni(
+    JNIEnv* env, jobject thiz, jlong native_handle) {
+  delete reinterpret_cast<BertNLClassifier*>(native_handle);
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_BertNLClassifier_initJniWithByteBuffer(
+    JNIEnv* env, jclass thiz, jobject model_buffer) {
+  auto model = GetMappedFileBuffer(env, model_buffer);
+  tflite::support::StatusOr<std::unique_ptr<BertNLClassifier>> status =
+      BertNLClassifier::CreateFromBuffer(model.data(), model.size());
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when initializing Bert NLClassifier: %s",
+                   status.status().message().data());
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_BertNLClassifier_initJniWithFileDescriptor(
+    JNIEnv* env, jclass thiz, jint fd) {
+  tflite::support::StatusOr<std::unique_ptr<BertNLClassifier>> status =
+      BertNLClassifier::CreateFromFd(fd);
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when initializing Bert NLClassifier: %s",
+                   status.status().message().data());
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jobject JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_BertNLClassifier_classifyNative(
+    JNIEnv* env, jclass clazz, jlong native_handle, jstring text) {
+  return RunClassifier(env, native_handle, text);
+}
+
+}  // namespace
diff --git a/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni.cc b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni.cc
new file mode 100644
index 0000000..d2ace75
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni.cc
@@ -0,0 +1,135 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include "tensorflow/lite/kernels/kernel_util.h"
+#include "tensorflow/lite/op_resolver.h"
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+#include "tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.h"
+
+namespace tflite {
+namespace task {
+// To be provided by a link-time library
+extern std::unique_ptr<OpResolver> CreateOpResolver();
+
+}  // namespace task
+}  // namespace tflite
+
+namespace {
+
+using ::tflite::support::utils::kAssertionError;
+using ::tflite::support::utils::kInvalidPointer;
+using ::tflite::support::utils::GetMappedFileBuffer;
+using ::tflite::support::utils::JStringToString;
+using ::tflite::support::utils::ThrowException;
+using ::tflite::task::text::nlclassifier::NLClassifier;
+using ::tflite::task::text::nlclassifier::NLClassifierOptions;
+using ::tflite::task::text::nlclassifier::RunClassifier;
+
+
+NLClassifierOptions ConvertJavaNLClassifierOptions(
+    JNIEnv* env, jobject java_nl_classifier_options) {
+  jclass nl_classifier_options_class = env->FindClass(
+      "org/tensorflow/lite/task/text/nlclassifier/"
+      "NLClassifier$NLClassifierOptions");
+  jmethodID input_tensor_index_method_id =
+      env->GetMethodID(nl_classifier_options_class, "inputTensorIndex", "()I");
+  jmethodID output_score_tensor_index_method_id = env->GetMethodID(
+      nl_classifier_options_class, "outputScoreTensorIndex", "()I");
+  jmethodID output_label_tensor_index_method_id = env->GetMethodID(
+      nl_classifier_options_class, "outputLabelTensorIndex", "()I");
+  jmethodID input_tensor_name_method_id = env->GetMethodID(
+      nl_classifier_options_class, "inputTensorName", "()Ljava/lang/String;");
+  jmethodID output_score_tensor_name_method_id =
+      env->GetMethodID(nl_classifier_options_class, "outputScoreTensorName",
+                       "()Ljava/lang/String;");
+  jmethodID output_label_tensor_name_method_id =
+      env->GetMethodID(nl_classifier_options_class, "outputLabelTensorName",
+                       "()Ljava/lang/String;");
+
+  return {
+      .input_tensor_index = env->CallIntMethod(java_nl_classifier_options,
+                                               input_tensor_index_method_id),
+      .output_score_tensor_index = env->CallIntMethod(
+          java_nl_classifier_options, output_score_tensor_index_method_id),
+      .output_label_tensor_index = env->CallIntMethod(
+          java_nl_classifier_options, output_label_tensor_index_method_id),
+      .input_tensor_name = JStringToString(
+          env, (jstring)env->CallObjectMethod(java_nl_classifier_options,
+                                              input_tensor_name_method_id)),
+      .output_score_tensor_name = JStringToString(
+          env,
+          (jstring)env->CallObjectMethod(java_nl_classifier_options,
+                                         output_score_tensor_name_method_id)),
+      .output_label_tensor_name = JStringToString(
+          env,
+          (jstring)env->CallObjectMethod(java_nl_classifier_options,
+                                         output_label_tensor_name_method_id)),
+  };
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_NLClassifier_deinitJni(
+    JNIEnv* env, jobject thiz, jlong native_handle) {
+  delete reinterpret_cast<NLClassifier*>(native_handle);
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_NLClassifier_initJniWithByteBuffer(
+    JNIEnv* env, jclass thiz, jobject nl_classifier_options,
+    jobject model_buffer) {
+  auto model = GetMappedFileBuffer(env, model_buffer);
+  tflite::support::StatusOr<std::unique_ptr<NLClassifier>> status =
+      NLClassifier::CreateFromBufferAndOptions(
+          model.data(), model.size(),
+          ConvertJavaNLClassifierOptions(env, nl_classifier_options),
+          tflite::task::CreateOpResolver());
+
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when initializing NLClassifier: %s",
+                   status.status().message().data());
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_NLClassifier_initJniWithFileDescriptor(
+    JNIEnv* env, jclass thiz, jobject nl_classifier_options, jint fd) {
+  tflite::support::StatusOr<std::unique_ptr<NLClassifier>> status =
+      NLClassifier::CreateFromFdAndOptions(
+          fd, ConvertJavaNLClassifierOptions(env, nl_classifier_options),
+          tflite::task::CreateOpResolver());
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when initializing NLClassifier: %s",
+                   status.status().message().data());
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jobject JNICALL
+Java_org_tensorflow_lite_task_text_nlclassifier_NLClassifier_classifyNative(
+    JNIEnv* env, jclass thiz, jlong native_handle, jstring text) {
+  return RunClassifier(env, native_handle, text);
+}
+
+}  // namespace
diff --git a/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.cc b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.cc
new file mode 100644
index 0000000..c358bee
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.cc
@@ -0,0 +1,56 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include "tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+using ::tflite::support::utils::ConvertVectorToArrayList;
+using ::tflite::support::utils::JStringToString;
+using ::tflite::task::core::Category;
+using ::tflite::task::text::nlclassifier::NLClassifier;
+
+jobject RunClassifier(JNIEnv* env, jlong native_handle, jstring text) {
+  auto* nl_classifier = reinterpret_cast<NLClassifier*>(native_handle);
+
+  auto results = nl_classifier->Classify(JStringToString(env, text));
+  jclass category_class =
+      env->FindClass("org/tensorflow/lite/support/label/Category");
+  jmethodID category_init =
+      env->GetMethodID(category_class, "<init>", "(Ljava/lang/String;F)V");
+
+  return ConvertVectorToArrayList<Category>(
+      env, results,
+      [env, category_class, category_init](const Category& category) {
+        jstring class_name = env->NewStringUTF(category.class_name.data());
+        // Convert double to float as Java interface exposes float as scores.
+        jobject jcategory =
+            env->NewObject(category_class, category_init, class_name,
+                           static_cast<float>(category.score));
+        env->DeleteLocalRef(class_name);
+        return jcategory;
+      });
+}
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.h b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.h
new file mode 100644
index 0000000..2c59ab5
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/nlclassifier/nl_classifier_jni_utils.h
@@ -0,0 +1,33 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_JAVA_SRC_NATIVE_TAKS_TEXT_NLCLASSIFIER_NL_CLASSIFIER_JNI_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_JAVA_SRC_NATIVE_TAKS_TEXT_NLCLASSIFIER_NL_CLASSIFIER_JNI_UTILS_H_
+
+#include <jni.h>
+
+namespace tflite {
+namespace task {
+namespace text {
+namespace nlclassifier {
+
+jobject RunClassifier(JNIEnv* env, jlong native_handle, jstring text);
+
+}  // namespace nlclassifier
+}  // namespace text
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_JAVA_SRC_NATIVE_TAKS_TEXT_NLCLASSIFIER_NL_CLASSIFIER_JNI_UTILS_H_
diff --git a/tensorflow_lite_support/java/src/native/task/text/qa/BUILD b/tensorflow_lite_support/java/src/native/task/text/qa/BUILD
new file mode 100644
index 0000000..9753e32
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/qa/BUILD
@@ -0,0 +1,30 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files([
+    "bert_question_answerer_jni.cc",
+])
+
+tflite_jni_binary(
+    name = "libtask_text_jni.so",
+    srcs = [
+        "bert_question_answerer_jni.cc",
+    ],
+    linkscript = "//tensorflow_lite_support/java:default_version_script.lds",
+    deps = [
+        "//tensorflow_lite_support/cc/task/text/qa:bert_question_answerer",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+    ],
+)
+
+cc_library(
+    name = "bert_question_answerer_native",
+    srcs = [
+        ":libtask_text_jni.so",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/text/qa/bert_question_answerer_jni.cc b/tensorflow_lite_support/java/src/native/task/text/qa/bert_question_answerer_jni.cc
new file mode 100644
index 0000000..92f9346
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/text/qa/bert_question_answerer_jni.cc
@@ -0,0 +1,127 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include "tensorflow_lite_support/cc/task/text/qa/bert_question_answerer.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+
+namespace {
+
+using ::tflite::support::utils::ConvertVectorToArrayList;
+using ::tflite::support::utils::GetMappedFileBuffer;
+using ::tflite::support::utils::JStringToString;
+using ::tflite::task::text::qa::BertQuestionAnswerer;
+using ::tflite::task::text::qa::QaAnswer;
+using ::tflite::task::text::qa::QuestionAnswerer;
+
+constexpr int kInvalidPointer = 0;
+
+extern "C" JNIEXPORT void JNICALL
+Java_org_tensorflow_lite_task_text_qa_BertQuestionAnswerer_deinitJni(
+    JNIEnv* env, jobject thiz, jlong native_handle) {
+  delete reinterpret_cast<QuestionAnswerer*>(native_handle);
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_qa_BertQuestionAnswerer_initJniWithModelWithMetadataByteBuffers(
+    JNIEnv* env, jclass thiz, jobjectArray model_buffers) {
+  absl::string_view model_with_metadata =
+      GetMappedFileBuffer(env, env->GetObjectArrayElement(model_buffers, 0));
+
+  tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>> status =
+      BertQuestionAnswerer::CreateFromBuffer(
+          model_with_metadata.data(), model_with_metadata.size());
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_qa_BertQuestionAnswerer_initJniWithFileDescriptor(
+    JNIEnv* env, jclass thiz, jint fd) {
+  tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>> status =
+      BertQuestionAnswerer::CreateFromFd(fd);
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_qa_BertQuestionAnswerer_initJniWithBertByteBuffers(
+    JNIEnv* env, jclass thiz, jobjectArray model_buffers) {
+  absl::string_view model =
+      GetMappedFileBuffer(env, env->GetObjectArrayElement(model_buffers, 0));
+  absl::string_view vocab =
+      GetMappedFileBuffer(env, env->GetObjectArrayElement(model_buffers, 1));
+
+  tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>> status =
+      BertQuestionAnswerer::CreateBertQuestionAnswererFromBuffer(
+          model.data(), model.size(), vocab.data(), vocab.size());
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_text_qa_BertQuestionAnswerer_initJniWithAlbertByteBuffers(
+    JNIEnv* env, jclass thiz, jobjectArray model_buffers) {
+  absl::string_view model =
+      GetMappedFileBuffer(env, env->GetObjectArrayElement(model_buffers, 0));
+  absl::string_view sp_model =
+      GetMappedFileBuffer(env, env->GetObjectArrayElement(model_buffers, 1));
+
+  tflite::support::StatusOr<std::unique_ptr<QuestionAnswerer>> status =
+      BertQuestionAnswerer::CreateAlbertQuestionAnswererFromBuffer(
+          model.data(), model.size(), sp_model.data(), sp_model.size());
+  if (status.ok()) {
+    return reinterpret_cast<jlong>(status->release());
+  } else {
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT jobject JNICALL
+Java_org_tensorflow_lite_task_text_qa_BertQuestionAnswerer_answerNative(
+    JNIEnv* env, jclass thiz, jlong native_handle, jstring context,
+    jstring question) {
+  auto* question_answerer = reinterpret_cast<QuestionAnswerer*>(native_handle);
+
+  std::vector<QaAnswer> results = question_answerer->Answer(
+      JStringToString(env, context), JStringToString(env, question));
+  jclass qa_answer_class =
+      env->FindClass("org/tensorflow/lite/task/text/qa/QaAnswer");
+  jmethodID qa_answer_ctor =
+      env->GetMethodID(qa_answer_class, "<init>", "(Ljava/lang/String;IIF)V");
+
+  return ConvertVectorToArrayList<QaAnswer>(
+      env, results,
+      [env, qa_answer_class, qa_answer_ctor](const QaAnswer& ans) {
+        jstring text = env->NewStringUTF(ans.text.data());
+        jobject qa_answer =
+            env->NewObject(qa_answer_class, qa_answer_ctor, text, ans.pos.start,
+                           ans.pos.end, ans.pos.logit);
+        env->DeleteLocalRef(text);
+        return qa_answer;
+      });
+}
+
+}  // namespace
diff --git a/tensorflow_lite_support/java/src/native/task/vision/BUILD b/tensorflow_lite_support/java/src/native/task/vision/BUILD
new file mode 100644
index 0000000..451a50c
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/BUILD
@@ -0,0 +1,59 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+cc_library(
+    name = "jni_utils",
+    srcs = [
+        "jni_utils.cc",
+    ],
+    hdrs = [
+        "jni_utils.h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/proto:class_proto_inc",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "@com_google_absl//absl/strings",
+    ],
+)
+
+cc_library(
+    name = "task_vision_native",
+    srcs = [
+        ":libtask_vision_jni.so",
+    ],
+)
+
+tflite_jni_binary(
+    name = "libtask_vision_jni.so",
+    srcs = [
+        "//tensorflow_lite_support/java/src/native/task/vision/classifier:image_classifier_jni.cc",
+        "//tensorflow_lite_support/java/src/native/task/vision/detector:object_detector_jni.cc",
+        "//tensorflow_lite_support/java/src/native/task/vision/segmenter:image_segmenter_jni.cc",
+    ],
+    linkscript = "//tensorflow_lite_support/java:default_version_script.lds",
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision:image_classifier",
+        "//tensorflow_lite_support/cc/task/vision:image_segmenter",
+        "//tensorflow_lite_support/cc/task/vision:object_detector",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:classifications_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:detections_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_classifier_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_segmenter_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:object_detector_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:segmentations_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "//tensorflow_lite_support/java/src/native/task/vision:jni_utils",
+        "@com_google_absl//absl/strings",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/vision/classifier/BUILD b/tensorflow_lite_support/java/src/native/task/vision/classifier/BUILD
new file mode 100644
index 0000000..8bddc2a
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/classifier/BUILD
@@ -0,0 +1,35 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files(["image_classifier_jni.cc"])
+
+cc_library(
+    name = "image_classifier_native",
+    srcs = [
+        ":libtask_vision_jni.so",
+    ],
+)
+
+tflite_jni_binary(
+    name = "libtask_vision_jni.so",
+    srcs = [
+        "image_classifier_jni.cc",
+    ],
+    linkscript = "//tensorflow_lite_support/java:default_version_script.lds",
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision:image_classifier",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:classifications_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_classifier_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "//tensorflow_lite_support/java/src/native/task/vision:jni_utils",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/vision/classifier/image_classifier_jni.cc b/tensorflow_lite_support/java/src/native/task/vision/classifier/image_classifier_jni.cc
new file mode 100644
index 0000000..2d52f93
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/classifier/image_classifier_jni.cc
@@ -0,0 +1,234 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include <memory>
+#include <string>
+
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/image_classifier.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/classifications_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/image_classifier_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+#include "tensorflow_lite_support/java/src/native/task/vision/jni_utils.h"
+
+namespace {
+
+using ::tflite::support::StatusOr;
+using ::tflite::support::utils::GetMappedFileBuffer;
+using ::tflite::support::utils::kAssertionError;
+using ::tflite::support::utils::kInvalidPointer;
+using ::tflite::support::utils::StringListToVector;
+using ::tflite::support::utils::ThrowException;
+using ::tflite::task::vision::BoundingBox;
+using ::tflite::task::vision::ClassificationResult;
+using ::tflite::task::vision::Classifications;
+using ::tflite::task::vision::ConvertToCategory;
+using ::tflite::task::vision::ConvertToFrameBufferOrientation;
+using ::tflite::task::vision::FrameBuffer;
+using ::tflite::task::vision::ImageClassifier;
+using ::tflite::task::vision::ImageClassifierOptions;
+
+// Creates an ImageClassifierOptions proto based on the Java class.
+ImageClassifierOptions ConvertToProtoOptions(JNIEnv* env,
+                                             jobject java_options) {
+  ImageClassifierOptions proto_options;
+  jclass java_options_class = env->FindClass(
+      "org/tensorflow/lite/task/vision/classifier/"
+      "ImageClassifier$ImageClassifierOptions");
+
+  jmethodID display_names_locale_id = env->GetMethodID(
+      java_options_class, "getDisplayNamesLocale", "()Ljava/lang/String;");
+  jstring display_names_locale = static_cast<jstring>(
+      env->CallObjectMethod(java_options, display_names_locale_id));
+  const char* pchars = env->GetStringUTFChars(display_names_locale, nullptr);
+  proto_options.set_display_names_locale(pchars);
+  env->ReleaseStringUTFChars(display_names_locale, pchars);
+
+  jmethodID max_results_id =
+      env->GetMethodID(java_options_class, "getMaxResults", "()I");
+  jint max_results = env->CallIntMethod(java_options, max_results_id);
+  proto_options.set_max_results(max_results);
+
+  jmethodID is_score_threshold_set_id =
+      env->GetMethodID(java_options_class, "getIsScoreThresholdSet", "()Z");
+  jboolean is_score_threshold_set =
+      env->CallBooleanMethod(java_options, is_score_threshold_set_id);
+  if (is_score_threshold_set) {
+    jmethodID score_threshold_id =
+        env->GetMethodID(java_options_class, "getScoreThreshold", "()F");
+    jfloat score_threshold =
+        env->CallFloatMethod(java_options, score_threshold_id);
+    proto_options.set_score_threshold(score_threshold);
+  }
+
+  jmethodID allow_list_id = env->GetMethodID(
+      java_options_class, "getLabelAllowList", "()Ljava/util/List;");
+  jobject allow_list = env->CallObjectMethod(java_options, allow_list_id);
+  auto allow_list_vector = StringListToVector(env, allow_list);
+  for (const auto& class_name : allow_list_vector) {
+    proto_options.add_class_name_whitelist(class_name);
+  }
+
+  jmethodID deny_list_id = env->GetMethodID(
+      java_options_class, "getLabelDenyList", "()Ljava/util/List;");
+  jobject deny_list = env->CallObjectMethod(java_options, deny_list_id);
+  auto deny_list_vector = StringListToVector(env, deny_list);
+  for (const auto& class_name : deny_list_vector) {
+    proto_options.add_class_name_blacklist(class_name);
+  }
+
+  jmethodID num_threads_id =
+      env->GetMethodID(java_options_class, "getNumThreads", "()I");
+  jint num_threads = env->CallIntMethod(java_options, num_threads_id);
+  proto_options.set_num_threads(num_threads);
+
+  return proto_options;
+}
+
+jobject ConvertToClassificationResults(JNIEnv* env,
+                                       const ClassificationResult& results) {
+  // jclass and init of Classifications.
+  jclass classifications_class = env->FindClass(
+      "org/tensorflow/lite/task/vision/classifier/Classifications");
+  jmethodID classifications_create =
+      env->GetStaticMethodID(classifications_class, "create",
+                             "(Ljava/util/List;I)Lorg/tensorflow/lite/"
+                             "task/vision/classifier/Classifications;");
+
+  // jclass, init, and add of ArrayList.
+  jclass array_list_class = env->FindClass("java/util/ArrayList");
+  jmethodID array_list_init =
+      env->GetMethodID(array_list_class, "<init>", "(I)V");
+  jmethodID array_list_add_method =
+      env->GetMethodID(array_list_class, "add", "(Ljava/lang/Object;)Z");
+
+  jobject classifications_list =
+      env->NewObject(array_list_class, array_list_init,
+                     static_cast<jint>(results.classifications_size()));
+  for (int i = 0; i < results.classifications_size(); i++) {
+    auto classifications = results.classifications(i);
+    jobject jcategory_list = env->NewObject(array_list_class, array_list_init,
+                                            classifications.classes_size());
+    for (const auto& classification : classifications.classes()) {
+      jobject jcategory = ConvertToCategory(env, classification);
+      env->CallBooleanMethod(jcategory_list, array_list_add_method, jcategory);
+
+      env->DeleteLocalRef(jcategory);
+    }
+    jobject jclassifications = env->CallStaticObjectMethod(
+        classifications_class, classifications_create, jcategory_list,
+        classifications.head_index());
+    env->CallBooleanMethod(classifications_list, array_list_add_method,
+                           jclassifications);
+
+    env->DeleteLocalRef(jcategory_list);
+    env->DeleteLocalRef(jclassifications);
+  }
+  return classifications_list;
+}
+
+jlong CreateImageClassifierFromOptions(JNIEnv* env,
+                                       const ImageClassifierOptions& options) {
+  StatusOr<std::unique_ptr<ImageClassifier>> image_classifier_or =
+      ImageClassifier::CreateFromOptions(options);
+  if (image_classifier_or.ok()) {
+    // Deletion is handled at deinitJni time.
+    return reinterpret_cast<jlong>(image_classifier_or->release());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when initializing ImageClassifier: %s",
+                   image_classifier_or.status().message().data());
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_org_tensorflow_lite_task_vision_classifier_ImageClassifier_deinitJni(
+    JNIEnv* env, jobject thiz, jlong native_handle) {
+  delete reinterpret_cast<ImageClassifier*>(native_handle);
+}
+
+// Creates an ImageClassifier instance from the model file descriptor.
+// file_descriptor_length and file_descriptor_offset are optional. Non-possitive
+// values will be ignored.
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_vision_classifier_ImageClassifier_initJniWithModelFdAndOptions(
+    JNIEnv* env, jclass thiz, jint file_descriptor,
+    jlong file_descriptor_length, jlong file_descriptor_offset,
+    jobject java_options) {
+  ImageClassifierOptions proto_options =
+      ConvertToProtoOptions(env, java_options);
+  auto file_descriptor_meta = proto_options.mutable_model_file_with_metadata()
+                                  ->mutable_file_descriptor_meta();
+  file_descriptor_meta->set_fd(file_descriptor);
+  if (file_descriptor_length > 0) {
+    file_descriptor_meta->set_length(file_descriptor_length);
+  }
+  if (file_descriptor_offset > 0) {
+    file_descriptor_meta->set_offset(file_descriptor_offset);
+  }
+  return CreateImageClassifierFromOptions(env, proto_options);
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_vision_classifier_ImageClassifier_initJniWithByteBuffer(
+    JNIEnv* env, jclass thiz, jobject model_buffer, jobject java_options) {
+  ImageClassifierOptions proto_options =
+      ConvertToProtoOptions(env, java_options);
+  // External proto generated header does not overload `set_file_content` with
+  // string_view, therefore GetMappedFileBuffer does not apply here.
+  // Creating a std::string will cause one extra copying of data. Thus, the
+  // most efficient way here is to set file_content using char* and its size.
+  proto_options.mutable_model_file_with_metadata()->set_file_content(
+      static_cast<char*>(env->GetDirectBufferAddress(model_buffer)),
+      static_cast<size_t>(env->GetDirectBufferCapacity(model_buffer)));
+  return CreateImageClassifierFromOptions(env, proto_options);
+}
+
+extern "C" JNIEXPORT jobject JNICALL
+Java_org_tensorflow_lite_task_vision_classifier_ImageClassifier_classifyNative(
+    JNIEnv* env, jclass thiz, jlong native_handle, jobject image_byte_buffer,
+    jint width, jint height, jintArray jroi, jint jorientation) {
+  auto* classifier = reinterpret_cast<ImageClassifier*>(native_handle);
+  auto image = GetMappedFileBuffer(env, image_byte_buffer);
+  std::unique_ptr<FrameBuffer> frame_buffer = CreateFromRgbRawBuffer(
+      reinterpret_cast<const uint8*>(image.data()),
+      FrameBuffer::Dimension{width, height},
+      ConvertToFrameBufferOrientation(env, jorientation));
+
+  int* roi_array = env->GetIntArrayElements(jroi, 0);
+  BoundingBox roi;
+  roi.set_origin_x(roi_array[0]);
+  roi.set_origin_y(roi_array[1]);
+  roi.set_width(roi_array[2]);
+  roi.set_height(roi_array[3]);
+  env->ReleaseIntArrayElements(jroi, roi_array, 0);
+
+  auto results_or = classifier->Classify(*frame_buffer, roi);
+  if (results_or.ok()) {
+    return ConvertToClassificationResults(env, results_or.value());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when classifying the image: %s",
+                   results_or.status().message().data());
+    return nullptr;
+  }
+}
+}  // namespace
diff --git a/tensorflow_lite_support/java/src/native/task/vision/detector/BUILD b/tensorflow_lite_support/java/src/native/task/vision/detector/BUILD
new file mode 100644
index 0000000..5abd3f1
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/detector/BUILD
@@ -0,0 +1,36 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files(["object_detector_jni.cc"])
+
+cc_library(
+    name = "object_detector_native",
+    srcs = [
+        ":libtask_vision_jni.so",
+    ],
+)
+
+tflite_jni_binary(
+    name = "libtask_vision_jni.so",
+    srcs = [
+        "object_detector_jni.cc",
+    ],
+    linkscript = "//tensorflow_lite_support/java:default_version_script.lds",
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision:object_detector",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/proto:bounding_box_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:detections_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:object_detector_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "//tensorflow_lite_support/java/src/native/task/vision:jni_utils",
+        "@com_google_absl//absl/strings",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/vision/detector/object_detector_jni.cc b/tensorflow_lite_support/java/src/native/task/vision/detector/object_detector_jni.cc
new file mode 100644
index 0000000..016b0bf
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/detector/object_detector_jni.cc
@@ -0,0 +1,228 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include <memory>
+#include <string>
+
+#include "absl/strings/string_view.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/object_detector.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/bounding_box_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/detections_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/object_detector_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+#include "tensorflow_lite_support/java/src/native/task/vision/jni_utils.h"
+
+namespace {
+
+using ::tflite::support::StatusOr;
+using ::tflite::support::utils::GetMappedFileBuffer;
+using ::tflite::support::utils::kAssertionError;
+using ::tflite::support::utils::kInvalidPointer;
+using ::tflite::support::utils::StringListToVector;
+using ::tflite::support::utils::ThrowException;
+using ::tflite::task::vision::BoundingBox;
+using ::tflite::task::vision::ConvertToCategory;
+using ::tflite::task::vision::ConvertToFrameBufferOrientation;
+using ::tflite::task::vision::DetectionResult;
+using ::tflite::task::vision::FrameBuffer;
+using ::tflite::task::vision::ObjectDetector;
+using ::tflite::task::vision::ObjectDetectorOptions;
+
+// Creates an ObjectDetectorOptions proto based on the Java class.
+ObjectDetectorOptions ConvertToProtoOptions(JNIEnv* env, jobject java_options) {
+  ObjectDetectorOptions proto_options;
+  jclass java_options_class = env->FindClass(
+      "org/tensorflow/lite/task/vision/detector/"
+      "ObjectDetector$ObjectDetectorOptions");
+
+  jmethodID display_names_locale_id = env->GetMethodID(
+      java_options_class, "getDisplayNamesLocale", "()Ljava/lang/String;");
+  jstring display_names_locale = static_cast<jstring>(
+      env->CallObjectMethod(java_options, display_names_locale_id));
+  const char* pchars = env->GetStringUTFChars(display_names_locale, nullptr);
+  proto_options.set_display_names_locale(pchars);
+  env->ReleaseStringUTFChars(display_names_locale, pchars);
+
+  jmethodID max_results_id =
+      env->GetMethodID(java_options_class, "getMaxResults", "()I");
+  jint max_results = env->CallIntMethod(java_options, max_results_id);
+  proto_options.set_max_results(max_results);
+
+  jmethodID is_score_threshold_set_id =
+      env->GetMethodID(java_options_class, "getIsScoreThresholdSet", "()Z");
+  jboolean is_score_threshold_set =
+      env->CallBooleanMethod(java_options, is_score_threshold_set_id);
+  if (is_score_threshold_set) {
+    jmethodID score_threshold_id =
+        env->GetMethodID(java_options_class, "getScoreThreshold", "()F");
+    jfloat score_threshold =
+        env->CallFloatMethod(java_options, score_threshold_id);
+    proto_options.set_score_threshold(score_threshold);
+  }
+
+  jmethodID allow_list_id = env->GetMethodID(
+      java_options_class, "getLabelAllowList", "()Ljava/util/List;");
+  jobject allow_list = env->CallObjectMethod(java_options, allow_list_id);
+  std::vector<std::string> allow_list_vector =
+      StringListToVector(env, allow_list);
+  for (const auto& class_name : allow_list_vector) {
+    proto_options.add_class_name_whitelist(class_name);
+  }
+
+  jmethodID deny_list_id = env->GetMethodID(
+      java_options_class, "getLabelDenyList", "()Ljava/util/List;");
+  jobject deny_list = env->CallObjectMethod(java_options, deny_list_id);
+  auto deny_list_vector = StringListToVector(env, deny_list);
+  for (const auto& class_name : deny_list_vector) {
+    proto_options.add_class_name_blacklist(class_name);
+  }
+
+  jmethodID num_threads_id =
+      env->GetMethodID(java_options_class, "getNumThreads", "()I");
+  jint num_threads = env->CallIntMethod(java_options, num_threads_id);
+  proto_options.set_num_threads(num_threads);
+
+  return proto_options;
+}
+
+jobject ConvertToDetectionResults(JNIEnv* env, const DetectionResult& results) {
+  // jclass and init of Detection.
+  jclass detection_class =
+      env->FindClass("org/tensorflow/lite/task/vision/detector/Detection");
+  jmethodID detection_create = env->GetStaticMethodID(
+      detection_class, "create",
+      "(Landroid/graphics/RectF;Ljava/util/List;)Lorg/tensorflow/lite/"
+      "task/vision/detector/Detection;");
+
+  // jclass, init, and add of ArrayList.
+  jclass array_list_class = env->FindClass("java/util/ArrayList");
+  jmethodID array_list_init =
+      env->GetMethodID(array_list_class, "<init>", "(I)V");
+  jmethodID array_list_add_method =
+      env->GetMethodID(array_list_class, "add", "(Ljava/lang/Object;)Z");
+
+  // jclass, init of RectF.
+  jclass rectf_class = env->FindClass("android/graphics/RectF");
+  jmethodID rectf_init = env->GetMethodID(rectf_class, "<init>", "(FFFF)V");
+
+  jobject detections_list =
+      env->NewObject(array_list_class, array_list_init,
+                     static_cast<jint>(results.detections_size()));
+
+  for (const auto& detection : results.detections()) {
+    // Create the category list.
+    jobject category_list = env->NewObject(array_list_class, array_list_init,
+                                           detection.classes_size());
+    for (const auto& classification : detection.classes()) {
+      jobject jcategory = ConvertToCategory(env, classification);
+      env->CallBooleanMethod(category_list, array_list_add_method, jcategory);
+    }
+
+    // Create the bounding box object.
+    const BoundingBox& bounding_box = detection.bounding_box();
+    float left = static_cast<float>(bounding_box.origin_x());
+    float top = static_cast<float>(bounding_box.origin_y());
+    float right = static_cast<float>(left + bounding_box.width());
+    float bottom = static_cast<float>(top + bounding_box.height());
+    jobject jbounding_box =
+        env->NewObject(rectf_class, rectf_init, left, top, right, bottom);
+
+    // Create the java Detection object.
+    jobject jdetection = env->CallStaticObjectMethod(
+        detection_class, detection_create, jbounding_box, category_list);
+    env->CallBooleanMethod(detections_list, array_list_add_method, jdetection);
+  }
+  return detections_list;
+}
+
+jlong CreateObjectDetectorFromOptions(JNIEnv* env,
+                                      const ObjectDetectorOptions& options) {
+  StatusOr<std::unique_ptr<ObjectDetector>> object_detector_or =
+      ObjectDetector::CreateFromOptions(options);
+  if (object_detector_or.ok()) {
+    return reinterpret_cast<jlong>(object_detector_or->release());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when initializing ObjectDetector: %s",
+                   object_detector_or.status().message().data());
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_org_tensorflow_lite_task_vision_detector_ObjectDetector_deinitJni(
+    JNIEnv* env, jobject thiz, jlong native_handle) {
+  delete reinterpret_cast<ObjectDetector*>(native_handle);
+}
+
+// Creates an ObjectDetector instance from the model file descriptor.
+// file_descriptor_length and file_descriptor_offset are optional. Non-possitive
+// values will be ignored.
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_vision_detector_ObjectDetector_initJniWithModelFdAndOptions(
+    JNIEnv* env, jclass thiz, jint file_descriptor,
+    jlong file_descriptor_length, jlong file_descriptor_offset,
+    jobject java_options) {
+  ObjectDetectorOptions proto_options =
+      ConvertToProtoOptions(env, java_options);
+  auto file_descriptor_meta = proto_options.mutable_model_file_with_metadata()
+                                  ->mutable_file_descriptor_meta();
+  file_descriptor_meta->set_fd(file_descriptor);
+  if (file_descriptor_length > 0) {
+    file_descriptor_meta->set_length(file_descriptor_length);
+  }
+  if (file_descriptor_offset > 0) {
+    file_descriptor_meta->set_offset(file_descriptor_offset);
+  }
+  return CreateObjectDetectorFromOptions(env, proto_options);
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_vision_detector_ObjectDetector_initJniWithByteBuffer(
+    JNIEnv* env, jclass thiz, jobject model_buffer, jobject java_options) {
+  ObjectDetectorOptions proto_options =
+      ConvertToProtoOptions(env, java_options);
+  proto_options.mutable_model_file_with_metadata()->set_file_content(
+      static_cast<char*>(env->GetDirectBufferAddress(model_buffer)),
+      static_cast<size_t>(env->GetDirectBufferCapacity(model_buffer)));
+  return CreateObjectDetectorFromOptions(env, proto_options);
+}
+
+extern "C" JNIEXPORT jobject JNICALL
+Java_org_tensorflow_lite_task_vision_detector_ObjectDetector_detectNative(
+    JNIEnv* env, jclass thiz, jlong native_handle, jobject image_byte_buffer,
+    jint width, jint height, jint jorientation) {
+  auto* detector = reinterpret_cast<ObjectDetector*>(native_handle);
+  absl::string_view image = GetMappedFileBuffer(env, image_byte_buffer);
+  std::unique_ptr<FrameBuffer> frame_buffer = CreateFromRgbRawBuffer(
+      reinterpret_cast<const uint8*>(image.data()),
+      FrameBuffer::Dimension{width, height},
+      ConvertToFrameBufferOrientation(env, jorientation));
+  auto results_or = detector->Detect(*frame_buffer);
+  if (results_or.ok()) {
+    return ConvertToDetectionResults(env, results_or.value());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when detecting the image: %s",
+                   results_or.status().message().data());
+    return nullptr;
+  }
+}
+}  // namespace
diff --git a/tensorflow_lite_support/java/src/native/task/vision/jni_utils.cc b/tensorflow_lite_support/java/src/native/task/vision/jni_utils.cc
new file mode 100644
index 0000000..af5dad9
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/jni_utils.cc
@@ -0,0 +1,85 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/java/src/native/task/vision/jni_utils.h"
+
+#include "absl/strings/str_cat.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+using ::tflite::support::utils::kAssertionError;
+using ::tflite::support::utils::ThrowException;
+
+constexpr char kCategoryClassName[] =
+    "org/tensorflow/lite/support/label/Category";
+constexpr char kStringClassName[] = "Ljava/lang/String;";
+constexpr char kEmptyString[] = "";
+
+jobject ConvertToCategory(JNIEnv* env, const Class& classification) {
+  // jclass and init of Category.
+  jclass category_class = env->FindClass(kCategoryClassName);
+  jmethodID category_create = env->GetStaticMethodID(
+      category_class, "create",
+      absl::StrCat("(", kStringClassName, kStringClassName, "F)L",
+                   kCategoryClassName, ";")
+          .c_str());
+
+  std::string label_string = classification.has_class_name()
+                                 ? classification.class_name()
+                                 : std::to_string(classification.index());
+  jstring label = env->NewStringUTF(label_string.c_str());
+  std::string display_name_string = classification.has_display_name()
+                                        ? classification.display_name()
+                                        : kEmptyString;
+  jstring display_name = env->NewStringUTF(display_name_string.c_str());
+  jobject jcategory =
+      env->CallStaticObjectMethod(category_class, category_create, label,
+                                  display_name, classification.score());
+  return jcategory;
+}
+
+FrameBuffer::Orientation ConvertToFrameBufferOrientation(JNIEnv* env,
+                                                         jint jorientation) {
+  switch (jorientation) {
+    case 0:
+      return FrameBuffer::Orientation::kTopLeft;
+    case 1:
+      return FrameBuffer::Orientation::kTopRight;
+    case 2:
+      return FrameBuffer::Orientation::kBottomRight;
+    case 3:
+      return FrameBuffer::Orientation::kBottomLeft;
+    case 4:
+      return FrameBuffer::Orientation::kLeftTop;
+    case 5:
+      return FrameBuffer::Orientation::kRightTop;
+    case 6:
+      return FrameBuffer::Orientation::kRightBottom;
+    case 7:
+      return FrameBuffer::Orientation::kLeftBottom;
+  }
+  // Should never happen.
+  ThrowException(env, kAssertionError,
+                 "The FrameBuffer Orientation type is unsupported: %d",
+                 jorientation);
+  return FrameBuffer::Orientation::kTopLeft;
+}
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
diff --git a/tensorflow_lite_support/java/src/native/task/vision/jni_utils.h b/tensorflow_lite_support/java/src/native/task/vision/jni_utils.h
new file mode 100644
index 0000000..7cb63f3
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/jni_utils.h
@@ -0,0 +1,38 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#ifndef TENSORFLOW_LITE_SUPPORT_JAVA_SRC_NATIVE_TASK_VISION_JNI_UTILS_H_
+#define TENSORFLOW_LITE_SUPPORT_JAVA_SRC_NATIVE_TASK_VISION_JNI_UTILS_H_
+
+#include <jni.h>
+
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/class_proto_inc.h"
+
+namespace tflite {
+namespace task {
+namespace vision {
+
+// Creates a Java Category object based on Class.
+jobject ConvertToCategory(JNIEnv* env, const Class& classification);
+
+FrameBuffer::Orientation ConvertToFrameBufferOrientation(JNIEnv* env,
+                                                         jint jorientation);
+
+}  // namespace vision
+}  // namespace task
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_JAVA_SRC_NATIVE_TASK_VISION_JNI_UTILS_H_
diff --git a/tensorflow_lite_support/java/src/native/task/vision/segmenter/BUILD b/tensorflow_lite_support/java/src/native/task/vision/segmenter/BUILD
new file mode 100644
index 0000000..23d601d
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/segmenter/BUILD
@@ -0,0 +1,34 @@
+load("@org_tensorflow//tensorflow/lite:build_def.bzl", "tflite_jni_binary")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files(["image_segmenter_jni.cc"])
+
+cc_library(
+    name = "image_segmenter_native",
+    srcs = [
+        ":libtask_vision_jni.so",
+    ],
+)
+
+tflite_jni_binary(
+    name = "libtask_vision_jni.so",
+    srcs = [
+        "image_segmenter_jni.cc",
+    ],
+    deps = [
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/cc/task/vision:image_segmenter",
+        "//tensorflow_lite_support/cc/task/vision/core:frame_buffer",
+        "//tensorflow_lite_support/cc/task/vision/proto:image_segmenter_options_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/proto:segmentations_proto_inc",
+        "//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
+        "//tensorflow_lite_support/cc/utils:jni_utils",
+        "//tensorflow_lite_support/java/jni",
+        "//tensorflow_lite_support/java/src/native/task/vision:jni_utils",
+        "@com_google_absl//absl/strings",
+    ],
+)
diff --git a/tensorflow_lite_support/java/src/native/task/vision/segmenter/image_segmenter_jni.cc b/tensorflow_lite_support/java/src/native/task/vision/segmenter/image_segmenter_jni.cc
new file mode 100644
index 0000000..1f6a2dc
--- /dev/null
+++ b/tensorflow_lite_support/java/src/native/task/vision/segmenter/image_segmenter_jni.cc
@@ -0,0 +1,238 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include <jni.h>
+
+#include <memory>
+#include <string>
+
+#include "absl/strings/str_cat.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/cc/task/vision/core/frame_buffer.h"
+#include "tensorflow_lite_support/cc/task/vision/image_segmenter.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/image_segmenter_options_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/proto/segmentations_proto_inc.h"
+#include "tensorflow_lite_support/cc/task/vision/utils/frame_buffer_common_utils.h"
+#include "tensorflow_lite_support/cc/utils/jni_utils.h"
+#include "tensorflow_lite_support/java/src/native/task/vision/jni_utils.h"
+
+namespace {
+
+using ::tflite::support::StatusOr;
+using ::tflite::support::utils::CreateByteArray;
+using ::tflite::support::utils::GetMappedFileBuffer;
+using ::tflite::support::utils::kAssertionError;
+using ::tflite::support::utils::kIllegalArgumentException;
+using ::tflite::support::utils::kInvalidPointer;
+using ::tflite::support::utils::ThrowException;
+using ::tflite::task::vision::ConvertToFrameBufferOrientation;
+using ::tflite::task::vision::FrameBuffer;
+using ::tflite::task::vision::ImageSegmenter;
+using ::tflite::task::vision::ImageSegmenterOptions;
+using ::tflite::task::vision::Segmentation;
+using ::tflite::task::vision::SegmentationResult;
+
+constexpr char kArrayListClassNameNoSig[] = "java/util/ArrayList";
+constexpr char kObjectClassName[] = "Ljava/lang/Object;";
+constexpr char kColorClassName[] = "Landroid/graphics/Color;";
+constexpr char kColorClassNameNoSig[] = "android/graphics/Color";
+constexpr char kColoredLabelClassName[] =
+    "Lorg/tensorflow/lite/task/vision/segmenter/ColoredLabel;";
+constexpr char kColoredLabelClassNameNoSig[] =
+    "org/tensorflow/lite/task/vision/segmenter/ColoredLabel";
+constexpr char kStringClassName[] = "Ljava/lang/String;";
+constexpr int kOutputTypeCategoryMask = 0;
+constexpr int kOutputTypeConfidenceMask = 1;
+
+// Creates an ImageSegmenterOptions proto based on the Java class.
+ImageSegmenterOptions ConvertToProtoOptions(JNIEnv* env,
+                                            jstring display_names_locale,
+                                            jint output_type,
+                                            jint num_threads) {
+  ImageSegmenterOptions proto_options;
+
+  const char* pchars = env->GetStringUTFChars(display_names_locale, nullptr);
+  proto_options.set_display_names_locale(pchars);
+  env->ReleaseStringUTFChars(display_names_locale, pchars);
+
+  switch (output_type) {
+    case kOutputTypeCategoryMask:
+      proto_options.set_output_type(ImageSegmenterOptions::CATEGORY_MASK);
+      break;
+    case kOutputTypeConfidenceMask:
+      proto_options.set_output_type(ImageSegmenterOptions::CONFIDENCE_MASK);
+      break;
+    default:
+      // Should never happen.
+      ThrowException(env, kIllegalArgumentException,
+                     "Unsupported output type: %d", output_type);
+  }
+
+  proto_options.set_num_threads(num_threads);
+
+  return proto_options;
+}
+
+void ConvertToSegmentationResults(JNIEnv* env,
+                                  const SegmentationResult& results,
+                                  jobject jmask_buffers, jintArray jmask_shape,
+                                  jobject jcolored_labels) {
+  if (results.segmentation_size() != 1) {
+    // Should never happen.
+    ThrowException(
+        env, kAssertionError,
+        "ImageSegmenter only supports one segmentation result, getting %d",
+        results.segmentation_size());
+  }
+
+  const Segmentation& segmentation = results.segmentation(0);
+
+  // Get the shape from the C++ Segmentation results.
+  int shape_array[2] = {segmentation.height(), segmentation.width()};
+  env->SetIntArrayRegion(jmask_shape, 0, 2, shape_array);
+
+  // jclass, init, and add of ArrayList.
+  jclass array_list_class = env->FindClass(kArrayListClassNameNoSig);
+  jmethodID array_list_add_method =
+      env->GetMethodID(array_list_class, "add",
+                       absl::StrCat("(", kObjectClassName, ")Z").c_str());
+
+  // Convert the masks into ByteBuffer list.
+  int num_pixels = segmentation.height() * segmentation.width();
+  if (segmentation.has_category_mask()) {
+    jbyteArray byte_array = CreateByteArray(
+        env,
+        reinterpret_cast<const jbyte*>(segmentation.category_mask().data()),
+        num_pixels * sizeof(uint8));
+    env->CallBooleanMethod(jmask_buffers, array_list_add_method, byte_array);
+    env->DeleteLocalRef(byte_array);
+  } else {
+    for (const auto& confidence_mask :
+         segmentation.confidence_masks().confidence_mask()) {
+      jbyteArray byte_array = CreateByteArray(
+          env, reinterpret_cast<const jbyte*>(confidence_mask.value().data()),
+          num_pixels * sizeof(float));
+      env->CallBooleanMethod(jmask_buffers, array_list_add_method, byte_array);
+      env->DeleteLocalRef(byte_array);
+    }
+  }
+
+  // Convert colored labels from the C++ object to the Java object.
+  jclass color_class = env->FindClass(kColorClassNameNoSig);
+  jmethodID color_rgb_method =
+      env->GetStaticMethodID(color_class, "rgb", "(III)I");
+  jclass colored_label_class = env->FindClass(kColoredLabelClassNameNoSig);
+  jmethodID colored_label_create_method = env->GetStaticMethodID(
+      colored_label_class, "create",
+      absl::StrCat("(", kStringClassName, kStringClassName, "I)",
+                   kColoredLabelClassName)
+          .c_str());
+
+  for (const auto& colored_label : segmentation.colored_labels()) {
+    jstring label = env->NewStringUTF(colored_label.class_name().c_str());
+    jstring display_name =
+        env->NewStringUTF(colored_label.display_name().c_str());
+    jint rgb = env->CallStaticIntMethod(color_class, color_rgb_method,
+                                        colored_label.r(), colored_label.g(),
+                                        colored_label.b());
+    jobject jcolored_label = env->CallStaticObjectMethod(
+        colored_label_class, colored_label_create_method, label, display_name,
+        rgb);
+    env->CallBooleanMethod(jcolored_labels, array_list_add_method,
+                           jcolored_label);
+
+    env->DeleteLocalRef(label);
+    env->DeleteLocalRef(display_name);
+    env->DeleteLocalRef(jcolored_label);
+  }
+}
+
+jlong CreateImageClassifierFromOptions(JNIEnv* env,
+                                       const ImageSegmenterOptions& options) {
+  StatusOr<std::unique_ptr<ImageSegmenter>> image_segmenter_or =
+      ImageSegmenter::CreateFromOptions(options);
+  if (image_segmenter_or.ok()) {
+    return reinterpret_cast<jlong>(image_segmenter_or->release());
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when initializing ImageSegmenter: %s",
+                   image_segmenter_or.status().message().data());
+    return kInvalidPointer;
+  }
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_org_tensorflow_lite_task_vision_segmenter_ImageSegmenter_deinitJni(
+    JNIEnv* env, jobject thiz, jlong native_handle) {
+  delete reinterpret_cast<ImageSegmenter*>(native_handle);
+}
+
+// Creates an ImageSegmenter instance from the model file descriptor.
+// file_descriptor_length and file_descriptor_offset are optional. Non-possitive
+// values will be ignored.
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_vision_segmenter_ImageSegmenter_initJniWithModelFdAndOptions(
+    JNIEnv* env, jclass thiz, jint file_descriptor,
+    jlong file_descriptor_length, jlong file_descriptor_offset,
+    jstring display_names_locale, jint output_type, jint num_threads) {
+  ImageSegmenterOptions proto_options = ConvertToProtoOptions(
+      env, display_names_locale, output_type, num_threads);
+  auto file_descriptor_meta = proto_options.mutable_model_file_with_metadata()
+                                  ->mutable_file_descriptor_meta();
+  file_descriptor_meta->set_fd(file_descriptor);
+  if (file_descriptor_length > 0) {
+    file_descriptor_meta->set_length(file_descriptor_length);
+  }
+  if (file_descriptor_offset > 0) {
+    file_descriptor_meta->set_offset(file_descriptor_offset);
+  }
+  return CreateImageClassifierFromOptions(env, proto_options);
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_org_tensorflow_lite_task_vision_segmenter_ImageSegmenter_initJniWithByteBuffer(
+    JNIEnv* env, jclass thiz, jobject model_buffer,
+    jstring display_names_locale, jint output_type, jint num_threads) {
+  ImageSegmenterOptions proto_options = ConvertToProtoOptions(
+      env, display_names_locale, output_type, num_threads);
+  proto_options.mutable_model_file_with_metadata()->set_file_content(
+      static_cast<char*>(env->GetDirectBufferAddress(model_buffer)),
+      static_cast<size_t>(env->GetDirectBufferCapacity(model_buffer)));
+  return CreateImageClassifierFromOptions(env, proto_options);
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_org_tensorflow_lite_task_vision_segmenter_ImageSegmenter_segmentNative(
+    JNIEnv* env, jclass thiz, jlong native_handle, jobject jimage_byte_buffer,
+    jint width, jint height, jobject jmask_buffers, jintArray jmask_shape,
+    jobject jcolored_labels, jint jorientation) {
+  auto* segmenter = reinterpret_cast<ImageSegmenter*>(native_handle);
+  absl::string_view image = GetMappedFileBuffer(env, jimage_byte_buffer);
+  std::unique_ptr<FrameBuffer> frame_buffer = CreateFromRgbRawBuffer(
+      reinterpret_cast<const uint8*>(image.data()),
+      FrameBuffer::Dimension{width, height},
+      ConvertToFrameBufferOrientation(env, jorientation));
+  auto results_or = segmenter->Segment(*frame_buffer);
+  if (results_or.ok()) {
+    ConvertToSegmentationResults(env, results_or.value(), jmask_buffers,
+                                 jmask_shape, jcolored_labels);
+  } else {
+    ThrowException(env, kAssertionError,
+                   "Error occurred when segmenting the image: %s",
+                   results_or.status().message().data());
+  }
+}
+
+}  // namespace
diff --git a/tensorflow_lite_support/metadata/BUILD b/tensorflow_lite_support/metadata/BUILD
new file mode 100644
index 0000000..db69bd2
--- /dev/null
+++ b/tensorflow_lite_support/metadata/BUILD
@@ -0,0 +1,51 @@
+load("@flatbuffers//:build_defs.bzl", "flatbuffer_android_library", "flatbuffer_cc_library", "flatbuffer_java_library", "flatbuffer_py_library")
+
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+exports_files(["metadata_schema.fbs"])
+
+flatbuffer_py_library(
+    name = "schema_py",
+    srcs = ["@org_tensorflow//tensorflow/lite/schema:schema.fbs"],
+)
+
+# Generic schema for inference on device.
+flatbuffer_android_library(
+    name = "schema_fbs_android",
+    srcs = ["@org_tensorflow//tensorflow/lite/schema:schema.fbs"],
+    custom_package = "org.tensorflow.lite.schema",
+)
+
+flatbuffer_java_library(
+    name = "schema_fbs_java",
+    srcs = ["@org_tensorflow//tensorflow/lite/schema:schema.fbs"],
+    custom_package = "org.tensorflow.lite.schema",
+)
+
+# Generic schema for model metadata.
+flatbuffer_cc_library(
+    name = "metadata_schema_cc",
+    srcs = ["metadata_schema.fbs"],
+)
+
+flatbuffer_py_library(
+    name = "metadata_schema_py",
+    srcs = ["metadata_schema.fbs"],
+)
+
+flatbuffer_java_library(
+    name = "metadata_schema_java",
+    srcs = ["metadata_schema.fbs"],
+    custom_package = "org.tensorflow.lite.support.metadata.schema",
+)
+
+flatbuffer_android_library(
+    name = "metadata_schema_fbs_android",
+    srcs = ["metadata_schema.fbs"],
+    custom_package = "org.tensorflow.lite.support.metadata.schema",
+)
diff --git a/tensorflow_lite_support/metadata/README.md b/tensorflow_lite_support/metadata/README.md
new file mode 100644
index 0000000..ff7d25f
--- /dev/null
+++ b/tensorflow_lite_support/metadata/README.md
@@ -0,0 +1,15 @@
+# TensorFlow Lite Metadata and Android wrapper code generator
+
+Note: Both TensorFlow Lite Metadata and the Android wrapper code generator are
+in experimental (beta) phase.
+
+TensorFlow Lite metadata provides a structured framework for storing metadata
+to convey information for both the developer that will utilitised the model and
+code generators which can create wrapper around the model. For information on
+how to populate model metadata, please refer to the [TensorFlow Lite Metadata 
+documentation](https://www.tensorflow.org/lite/convert/metadata).
+
+The first code generator which takes advantage of this metadata format is the
+TensorFlow Lite Android Code Generator. For more information on how to use this
+generator, please refer to the [TensorFlow Lite Android wrapper code generator
+documentation](https://www.tensorflow.org/lite/guide/codegen).
diff --git a/tensorflow_lite_support/metadata/build_defs.bzl b/tensorflow_lite_support/metadata/build_defs.bzl
new file mode 100644
index 0000000..8bdab12
--- /dev/null
+++ b/tensorflow_lite_support/metadata/build_defs.bzl
@@ -0,0 +1,43 @@
+"""Build rules to generate metadata schema versions."""
+
+METADATA_SCHEMA_FILE = "//tensorflow_lite_support/metadata:metadata_schema.fbs"
+
+def stamp_metadata_parser_version(
+        name,
+        srcs,
+        outs):
+    """Stamps the latest metadata parser version into the srcs files.
+
+    Replaces all the occurrences of "{LATEST_METADATA_PARSER_VERSION}" in the
+    srcs files with the metadata schema version extracted from
+    METADATA_SCHEMA_FILE and then outputs the generated file into outs,
+    respectively. The number of srcs files needs to match the number of outs
+    files.
+
+    Args:
+        name: Rule name. (required)
+        srcs: List of source files. (required)
+        outs: List of output files. (required)
+    """
+    if len(srcs) != len(outs):
+        fail(("The number of srcs files (%d) does not match that of the outs" +
+              " files (%d).") %
+             (len(srcs), len(outs)))
+
+    for i in range(0, len(srcs)):
+        native.genrule(
+            name = "%s_file%d" % (name, i),
+            srcs = [srcs[i]],
+            outs = [outs[i]],
+            tools = [METADATA_SCHEMA_FILE],
+            # Gets the metadata schema version from the file, and stamps it
+            # into the srcs file.
+            cmd = "version=$$(sed -n -e '/Schema Semantic version/ s/.*\\: *//p' $(location %s));" %
+                  METADATA_SCHEMA_FILE +
+                  'sed "s/{LATEST_METADATA_PARSER_VERSION}/$$version/" $< > $@',
+        )
+
+    native.filegroup(
+        name = name,
+        srcs = outs,
+    )
diff --git a/tensorflow_lite_support/metadata/cc/BUILD b/tensorflow_lite_support/metadata/cc/BUILD
new file mode 100644
index 0000000..ed5bedc
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/BUILD
@@ -0,0 +1,53 @@
+load("//tensorflow_lite_support/metadata:build_defs.bzl", "stamp_metadata_parser_version")
+
+package(
+    default_visibility = ["//tensorflow_lite_support:users"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+stamp_metadata_parser_version(
+    name = "metadata_parser_h",
+    srcs = ["metadata_parser.h.template"],
+    outs = ["metadata_parser.h"],
+)
+
+cc_library(
+    name = "metadata_extractor",
+    srcs = ["metadata_extractor.cc"],
+    hdrs = ["metadata_extractor.h"],
+    deps = [
+        "@com_google_absl//absl/container:flat_hash_map",
+        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/strings:str_format",
+        "@flatbuffers",
+        "@org_libzip//:zip",
+    ] + select({
+        "//tensorflow_lite_support/cc:tflite_use_c_api": ["@org_tensorflow//tensorflow/lite/c:c_api"],
+        "//conditions:default": ["@org_tensorflow//tensorflow/lite:framework"],
+    }) + [
+        "@org_tensorflow//tensorflow/lite/schema:schema_fbs",
+        "//tensorflow_lite_support/cc:common",
+        "//tensorflow_lite_support/cc/port:status_macros",
+        "//tensorflow_lite_support/cc/port:statusor",
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+    ],
+)
+
+cc_library(
+    name = "metadata_version",
+    srcs = ["metadata_version.cc"],
+    hdrs = [
+        "metadata_version.h",
+        ":metadata_parser_h",
+    ],
+    deps = [
+        "//tensorflow_lite_support/metadata:metadata_schema_cc",
+        "@com_google_absl//absl/strings",
+        "@flatbuffers",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@org_tensorflow//tensorflow/lite/kernels/internal:compatibility",
+        "@org_tensorflow//tensorflow/lite/tools:logging",
+    ],
+)
diff --git a/tensorflow_lite_support/metadata/cc/metadata_extractor.cc b/tensorflow_lite_support/metadata/cc/metadata_extractor.cc
new file mode 100644
index 0000000..cf4edaa
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/metadata_extractor.cc
@@ -0,0 +1,366 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/metadata/cc/metadata_extractor.h"
+
+#include <functional>
+
+#include "absl/memory/memory.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_format.h"
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "lib/zip.h"  // from @org_libzip
+#include "tensorflow/lite/schema/schema_generated.h"
+#include "tensorflow_lite_support/cc/common.h"
+#include "tensorflow_lite_support/cc/port/status_macros.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+#if TFLITE_USE_C_API
+#include "tensorflow/lite/c/c_api.h"
+#else
+#include "tensorflow/lite/model_builder.h"
+#endif
+
+namespace tflite {
+namespace metadata {
+
+namespace {
+constexpr char kMetadataBufferName[] = "TFLITE_METADATA";
+
+using ::absl::StatusCode;
+using ::flatbuffers::Offset;
+using ::flatbuffers::Vector;
+using ::tflite::TensorMetadata;
+using ::tflite::support::CreateStatusWithPayload;
+using ::tflite::support::TfLiteSupportStatus;
+
+// Helper class that takes a callback function, and invokes it in its
+// destructor.
+class SimpleCleanUp {
+ public:
+  explicit SimpleCleanUp(std::function<void()> callback)
+      : callback_(std::move(callback)) {}
+
+  ~SimpleCleanUp() {
+    if (callback_ != nullptr) callback_();
+  }
+
+  // Use `std::move(simple_cleanup).Cancel()` to prevent the callback from ever
+  // executing at all. Once a SimpleCleanUp object has been `std::move(...)`-ed,
+  // it may not be read from again.
+  void Cancel() && { callback_ = nullptr; }
+
+ private:
+  std::function<void()> callback_;
+};
+
+// Util to get item from src_vector specified by index.
+template <typename T>
+const T* GetItemFromVector(
+    const flatbuffers::Vector<flatbuffers::Offset<T>>* src_vector, int index) {
+  if (src_vector == nullptr || index < 0 || index >= src_vector->size()) {
+    return nullptr;
+  }
+  return src_vector->Get(index);
+}
+}  // namespace
+
+/* static */
+tflite::support::StatusOr<std::unique_ptr<ModelMetadataExtractor>>
+ModelMetadataExtractor::CreateFromModelBuffer(const char* buffer_data,
+                                              size_t buffer_size) {
+  // Use absl::WrapUnique() to call private constructor:
+  // https://abseil.io/tips/126.
+  std::unique_ptr<ModelMetadataExtractor> extractor =
+      absl::WrapUnique(new ModelMetadataExtractor());
+  RETURN_IF_ERROR(extractor->InitFromModelBuffer(buffer_data, buffer_size));
+  return extractor;
+}
+
+/* static */
+tflite::support::StatusOr<const tflite::ProcessUnit*>
+ModelMetadataExtractor::FindFirstProcessUnit(
+    const tflite::TensorMetadata& tensor_metadata,
+    tflite::ProcessUnitOptions type) {
+  const tflite::ProcessUnit* result = nullptr;
+  if (tensor_metadata.process_units() == nullptr) {
+    return result;
+  }
+  for (const auto process_unit : *tensor_metadata.process_units()) {
+    if (process_unit->options_type() == type) {
+      if (result != nullptr) {
+        return CreateStatusWithPayload(
+            StatusCode::kInvalidArgument,
+            absl::StrCat("Found multiple ProcessUnits with type=",
+                         tflite::EnumNameProcessUnitOptions(type),
+                         ", expected at most one."),
+            TfLiteSupportStatus::kMetadataInvalidProcessUnitsError);
+      }
+      result = process_unit;
+    }
+  }
+  return result;
+}
+
+/* static */
+std::string ModelMetadataExtractor::FindFirstAssociatedFileName(
+    const tflite::TensorMetadata& tensor_metadata,
+    tflite::AssociatedFileType type, absl::string_view locale) {
+  if (tensor_metadata.associated_files() == nullptr) {
+    return std::string();
+  }
+  for (const auto associated_file : *tensor_metadata.associated_files()) {
+    if (associated_file->type() != type || associated_file->name() == nullptr) {
+      continue;
+    }
+    if (locale.empty() || (associated_file->locale() != nullptr &&
+                           locale == associated_file->locale()->str())) {
+      return associated_file->name()->str();
+    }
+  }
+  return std::string();
+}
+
+absl::Status ModelMetadataExtractor::InitFromModelBuffer(
+    const char* buffer_data, size_t buffer_size) {
+  // Rely on the simplest, base flatbuffers verifier. Here is not the place to
+  // e.g. use an OpResolver: we just want to make sure the buffer is valid to
+  // access the metadata.
+  flatbuffers::Verifier verifier = flatbuffers::Verifier(
+      reinterpret_cast<const uint8_t*>(buffer_data), buffer_size);
+  if (!tflite::VerifyModelBuffer(verifier)) {
+    return CreateStatusWithPayload(
+        StatusCode::kInvalidArgument,
+        "The model is not a valid FlatBuffer buffer.",
+        TfLiteSupportStatus::kInvalidFlatBufferError);
+  }
+  model_ = tflite::GetModel(buffer_data);
+  if (model_->metadata() == nullptr) {
+    // Not all models have metadata, which is OK. `GetModelMetadata()` then
+    // returns nullptr.
+    return absl::OkStatus();
+  }
+  // Look for the "TFLITE_METADATA" field, if any.
+  for (int i = 0; i < model_->metadata()->size(); ++i) {
+    const auto metadata = model_->metadata()->Get(i);
+    if (metadata->name()->str() != kMetadataBufferName) {
+      continue;
+    }
+    const auto buffer_index = metadata->buffer();
+    const auto metadata_buffer =
+        model_->buffers()->Get(buffer_index)->data()->data();
+    if (!tflite::ModelMetadataBufferHasIdentifier(metadata_buffer)) {
+      return CreateStatusWithPayload(
+          StatusCode::kInvalidArgument,
+          absl::StrFormat(
+              "Invalid metadata schema version: expected %s, got %s",
+              absl::string_view(tflite::ModelMetadataIdentifier())
+                  .substr(
+                      0, flatbuffers::FlatBufferBuilder::kFileIdentifierLength),
+              // Returned identifier is not null terminated; has to be
+              // truncated.
+              absl::string_view(
+                  flatbuffers::GetBufferIdentifier(metadata_buffer))
+                  .substr(
+                      0,
+                      flatbuffers::FlatBufferBuilder::kFileIdentifierLength)),
+          TfLiteSupportStatus::kMetadataInvalidSchemaVersionError);
+    }
+    model_metadata_ = tflite::GetModelMetadata(metadata_buffer);
+    if (model_metadata_ == nullptr) {
+      return CreateStatusWithPayload(StatusCode::kInternal,
+                                     "Expected Model Metadata not to be null.");
+    }
+    return ExtractAssociatedFiles(buffer_data, buffer_size);
+    break;
+  }
+  return absl::OkStatus();
+}
+
+absl::Status ModelMetadataExtractor::ExtractAssociatedFiles(
+    const char* buffer_data, size_t buffer_size) {
+  // Setup libzip error reporting.
+  zip_error_t error;
+  zip_error_init(&error);
+  auto zip_error_cleanup = SimpleCleanUp([&error] { zip_error_fini(&error); });
+
+  // Initialize zip source.
+  zip_source_t* src =
+      zip_source_buffer_create(buffer_data, buffer_size, /*freep=*/0, &error);
+  if (src == nullptr) {
+    return CreateStatusWithPayload(
+        StatusCode::kUnknown,
+        absl::StrFormat("Can't create zip source from model buffer: %s",
+                        zip_error_strerror(&error)),
+        TfLiteSupportStatus::kMetadataAssociatedFileZipError);
+  }
+  auto zip_source_cleanup = SimpleCleanUp([src] { zip_source_free(src); });
+
+  // Try opening zip source.
+  zip* zip_archive = zip_open_from_source(src, /*flags=*/0, &error);
+  if (zip_archive == nullptr) {
+    // It's OK if it fails: this means there are no associated files with this
+    // model.
+    return absl::OkStatus();
+  }
+  auto zip_archive_cleanup =
+      SimpleCleanUp([zip_archive] { zip_close(zip_archive); });
+  // As per the documentation [1] for zip_source_free, it should not be called
+  // after a successful call to zip_open_from_source.
+  //
+  // [1]: https://libzip.org/documentation/zip_source_free.html
+  std::move(zip_source_cleanup).Cancel();
+
+  const int num_files = zip_get_num_entries(zip_archive, /*flags=*/0);
+  for (int index = 0; index < num_files; ++index) {
+    // Get file stats.
+    struct zip_stat zip_file_stat;
+    zip_stat_init(&zip_file_stat);
+    zip_stat_index(zip_archive, index, /*flags=*/0, &zip_file_stat);
+    absl::string_view filename = zip_file_stat.name;
+    const auto unzip_filesize = zip_file_stat.size;
+
+    // Open file.
+    zip_file* zip_file = zip_fopen_index(zip_archive, index, /*flags=*/0);
+    if (zip_file == nullptr) {
+      return CreateStatusWithPayload(
+          StatusCode::kUnknown,
+          absl::StrFormat("Unable to open associated file with name: %s",
+                          zip_file_stat.name),
+          TfLiteSupportStatus::kMetadataAssociatedFileZipError);
+    }
+    auto zip_file_cleanup = SimpleCleanUp([zip_file] { zip_fclose(zip_file); });
+
+    // Unzip file.
+    char* unzip_buffer = new char[unzip_filesize];
+    auto unzip_buffer_cleanup =
+        SimpleCleanUp([unzip_buffer] { delete[] unzip_buffer; });
+    if (zip_fread(zip_file, unzip_buffer, unzip_filesize) != unzip_filesize) {
+      return CreateStatusWithPayload(
+          StatusCode::kUnknown,
+          absl::StrFormat("Unzipping failed for file: %s.", filename),
+          TfLiteSupportStatus::kMetadataAssociatedFileZipError);
+    }
+
+    // Copy file contents in map.
+    associated_files_[filename] = std::string(unzip_buffer, unzip_filesize);
+  }
+  return absl::OkStatus();
+}
+
+tflite::support::StatusOr<absl::string_view>
+ModelMetadataExtractor::GetAssociatedFile(const std::string& filename) const {
+  auto it = associated_files_.find(filename);
+  if (it == associated_files_.end()) {
+    return CreateStatusWithPayload(
+        StatusCode::kNotFound,
+        absl::StrFormat("No associated file with name: %s", filename),
+        TfLiteSupportStatus::kMetadataAssociatedFileNotFoundError);
+  }
+  return it->second;
+}
+
+const flatbuffers::Vector<flatbuffers::Offset<tflite::TensorMetadata>>*
+ModelMetadataExtractor::GetInputTensorMetadata() const {
+  if (model_metadata_ == nullptr ||
+      model_metadata_->subgraph_metadata() == nullptr) {
+    return nullptr;
+  }
+  return model_metadata_->subgraph_metadata()
+      ->Get(kDefaultSubgraphIndex)
+      ->input_tensor_metadata();
+}
+
+const tflite::TensorMetadata* ModelMetadataExtractor::GetInputTensorMetadata(
+    int index) const {
+  return GetItemFromVector<tflite::TensorMetadata>(GetInputTensorMetadata(),
+                                                   index);
+}
+
+int ModelMetadataExtractor::GetInputTensorCount() const {
+  const flatbuffers::Vector<flatbuffers::Offset<tflite::TensorMetadata>>*
+      input_tensor_metadata = GetInputTensorMetadata();
+  return input_tensor_metadata == nullptr ? 0 : input_tensor_metadata->size();
+}
+
+const Vector<Offset<TensorMetadata>>*
+ModelMetadataExtractor::GetOutputTensorMetadata() const {
+  if (model_metadata_ == nullptr ||
+      model_metadata_->subgraph_metadata() == nullptr) {
+    return nullptr;
+  }
+  return model_metadata_->subgraph_metadata()
+      ->Get(kDefaultSubgraphIndex)
+      ->output_tensor_metadata();
+}
+
+const tflite::TensorMetadata* ModelMetadataExtractor::GetOutputTensorMetadata(
+    int index) const {
+  return GetItemFromVector<tflite::TensorMetadata>(GetOutputTensorMetadata(),
+                                                   index);
+}
+
+int ModelMetadataExtractor::GetOutputTensorCount() const {
+  const flatbuffers::Vector<flatbuffers::Offset<tflite::TensorMetadata>>*
+      output_tensor_metadata = GetOutputTensorMetadata();
+  return output_tensor_metadata == nullptr ? 0 : output_tensor_metadata->size();
+}
+
+const Vector<flatbuffers::Offset<tflite::ProcessUnit>>*
+ModelMetadataExtractor::GetInputProcessUnits() const {
+  if (model_metadata_ == nullptr ||
+      model_metadata_->subgraph_metadata() == nullptr) {
+    return nullptr;
+  }
+  return model_metadata_->subgraph_metadata()
+      ->Get(kDefaultSubgraphIndex)
+      ->input_process_units();
+}
+
+const tflite::ProcessUnit* ModelMetadataExtractor::GetInputProcessUnit(
+    int index) const {
+  return GetItemFromVector<tflite::ProcessUnit>(GetInputProcessUnits(), index);
+}
+
+int ModelMetadataExtractor::GetInputProcessUnitsCount() const {
+  const Vector<flatbuffers::Offset<tflite::ProcessUnit>>* input_process_units =
+      GetInputProcessUnits();
+  return input_process_units == nullptr ? 0 : input_process_units->size();
+}
+
+const Vector<flatbuffers::Offset<tflite::ProcessUnit>>*
+ModelMetadataExtractor::GetOutputProcessUnits() const {
+  if (model_metadata_ == nullptr ||
+      model_metadata_->subgraph_metadata() == nullptr) {
+    return nullptr;
+  }
+  return model_metadata_->subgraph_metadata()
+      ->Get(kDefaultSubgraphIndex)
+      ->output_process_units();
+}
+
+const tflite::ProcessUnit* ModelMetadataExtractor::GetOutputProcessUnit(
+    int index) const {
+  return GetItemFromVector<tflite::ProcessUnit>(GetOutputProcessUnits(), index);
+}
+
+int ModelMetadataExtractor::GetOutputProcessUnitsCount() const {
+  const Vector<flatbuffers::Offset<tflite::ProcessUnit>>* output_process_units =
+      GetOutputProcessUnits();
+  return output_process_units == nullptr ? 0 : output_process_units->size();
+}
+
+}  // namespace metadata
+}  // namespace tflite
diff --git a/tensorflow_lite_support/metadata/cc/metadata_extractor.h b/tensorflow_lite_support/metadata/cc/metadata_extractor.h
new file mode 100644
index 0000000..8eafe93
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/metadata_extractor.h
@@ -0,0 +1,157 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_EXTRACTOR_H_
+#define TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_EXTRACTOR_H_
+
+#include "absl/container/flat_hash_map.h"
+#include "absl/status/status.h"
+#include "absl/strings/string_view.h"
+#include "tensorflow/lite/schema/schema_generated.h"
+#include "tensorflow_lite_support/cc/port/statusor.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace metadata {
+
+// Extracts and provides easy access to the TFLite ModelMetadata [1] and
+// corresponding associated files packed into a TFLite FlatBuffer, if any.
+//
+// [1]: https://www.tensorflow.org/lite/convert/metadata
+class ModelMetadataExtractor {
+ public:
+  // Creates a ModelMetadataExtractor from the provided TFLite Model FlatBuffer
+  // and returns a pointer to the new object. Ownership is transferred to the
+  // caller. Returns an error if the creation failed, which may happen e.g. if
+  // the provided buffer is not a valid TFLite FlatBuffer.
+  //
+  // Warning: Does not take ownership of the provided buffer, which must outlive
+  // this object.
+  //
+  // It is recommended to obtain and manage the buffer through an
+  // ExternalFileHandler[1], which is optimized through mmap(2) to avoid having
+  // to load the entire buffer in memory when provided by path or file
+  // descriptor.
+  //
+  // [1]:
+  // tensorflow_lite_support/c/task/core/external_file_handler.h
+  static tflite::support::StatusOr<std::unique_ptr<ModelMetadataExtractor>>
+  CreateFromModelBuffer(const char* buffer_data, size_t buffer_size);
+
+  // Returns the pointer to the *first* ProcessUnit with the provided type, or
+  // nullptr if none can be found. An error is returned if multiple
+  // ProcessUnit-s with the provided type are found.
+  static tflite::support::StatusOr<const tflite::ProcessUnit*>
+  FindFirstProcessUnit(const tflite::TensorMetadata& tensor_metadata,
+                       tflite::ProcessUnitOptions type);
+
+  // Returns the name of the *first* associated file with the provided type and
+  // (optional) locale in the provided TensorMetadata, or an empty string if no
+  // such associated file can be found (which is not necessarily an error: some
+  // models have no associated files at all) or its `name` field is unspecified.
+  // Note: see `GetAssociatedFile` to read the actual file contents.
+  static std::string FindFirstAssociatedFileName(
+      const tflite::TensorMetadata& tensor_metadata,
+      tflite::AssociatedFileType type,
+      absl::string_view locale = absl::string_view());
+
+  // Returns a pointer to the extracted TFLite Model Metadata, or nullptr if no
+  // metadata was present in the Model FlatBuffer provided at creation time.
+  const tflite::ModelMetadata* GetModelMetadata() const {
+    return model_metadata_;
+  }
+
+  // Gets the contents of the associated file with the provided name packed into
+  // the model metadata. An error is returned if there is no such associated
+  // file.
+  tflite::support::StatusOr<absl::string_view> GetAssociatedFile(
+      const std::string& filename) const;
+
+  // Note: all methods below retrieves metadata of the *first* subgraph as
+  // default.
+
+  // Gets the metadata for input tensors.
+  const flatbuffers::Vector<flatbuffers::Offset<tflite::TensorMetadata>>*
+  GetInputTensorMetadata() const;
+
+  // Gets the metadata for the input tensor specified by the given index, or
+  // nullptr in case there is no metadata or the index is out of range.
+  const tflite::TensorMetadata* GetInputTensorMetadata(int index) const;
+
+  // Gets the count of input tensors with metadata in the metadata FlatBuffer.
+  // In particular, 0 is returned when there is no metadata.
+  int GetInputTensorCount() const;
+
+  // Gets the metadata for output tensors.
+  const flatbuffers::Vector<flatbuffers::Offset<tflite::TensorMetadata>>*
+  GetOutputTensorMetadata() const;
+
+  // Gets the metadata for the output tensor specified by the given index, or
+  // nullptr in case there is no metadata or the index is out of range.
+  const tflite::TensorMetadata* GetOutputTensorMetadata(int index) const;
+
+  // Gets the count of output tensors with metadata in the metadata FlatBuffer.
+  // In particular, 0 is returned when there is no metadata.
+  int GetOutputTensorCount() const;
+
+  // Gets the input process units from SubgraphMetadata.input_process_units,
+  // could be nullptr.
+  const flatbuffers::Vector<flatbuffers::Offset<tflite::ProcessUnit>>*
+  GetInputProcessUnits() const;
+
+  // Gets the input process unit specified by the given index, or nullptr in
+  // case there is no input process unit or the index is out of range.
+  const tflite::ProcessUnit* GetInputProcessUnit(int index) const;
+
+  // Gets the count of input process units. In particular, 0 is returned when
+  // there is no input process units.
+  int GetInputProcessUnitsCount() const;
+
+  // Gets the output process units from SubgraphMetadata.output_process_units,
+  // could be nullptr.
+  const flatbuffers::Vector<flatbuffers::Offset<tflite::ProcessUnit>>*
+  GetOutputProcessUnits() const;
+
+  // Gets the output process unit specified by the given index, or nullptr in
+  // case there is no output process unit or the index is out of range.
+  const tflite::ProcessUnit* GetOutputProcessUnit(int index) const;
+
+  // Gets the count of output process units. In particular, 0 is returned when
+  // there is no output process units.
+  int GetOutputProcessUnitsCount() const;
+
+ private:
+  static constexpr int kDefaultSubgraphIndex = 0;
+  // Private default constructor, called from CreateFromModel().
+  ModelMetadataExtractor() = default;
+  // Initializes the ModelMetadataExtractor from the provided Model FlatBuffer.
+  absl::Status InitFromModelBuffer(const char* buffer_data, size_t buffer_size);
+  // Extracts and stores in associated_files_ the associated files (if present)
+  // packed into the model FlatBuffer data.
+  absl::Status ExtractAssociatedFiles(const char* buffer_data,
+                                      size_t buffer_size);
+  // Pointer to the TFLite Model object from which to read the ModelMetadata.
+  const tflite::Model* model_{nullptr};
+  // Pointer to the extracted ModelMetadata, if any.
+  const tflite::ModelMetadata* model_metadata_{nullptr};
+  // The files associated with the ModelMetadata, as a map with the filename
+  // (corresponding to a basename, e.g. "labels.txt") as key and the file
+  // contents as value.
+  absl::flat_hash_map<std::string, std::string> associated_files_;
+};
+
+}  // namespace metadata
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_EXTRACTOR_H_
diff --git a/tensorflow_lite_support/metadata/cc/metadata_parser.h.template b/tensorflow_lite_support/metadata/cc/metadata_parser.h.template
new file mode 100644
index 0000000..7e26050
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/metadata_parser.h.template
@@ -0,0 +1,28 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_PARSER_H_
+#define TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_PARSER_H_
+
+namespace tflite {
+namespace metadata {
+
+// The version of the metadata parser that this metadata versioning library is
+// depending on.
+inline constexpr char kMatadataParserVersion[] = "{LATEST_METADATA_PARSER_VERSION}";
+
+}  // namespace metadata
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_PARSER_H_
diff --git a/tensorflow_lite_support/metadata/cc/metadata_version.cc b/tensorflow_lite_support/metadata/cc/metadata_version.cc
new file mode 100644
index 0000000..7679f6c
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/metadata_version.cc
@@ -0,0 +1,302 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#include "tensorflow_lite_support/metadata/cc/metadata_version.h"
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include <array>
+#include <ostream>
+#include <string>
+#include <vector>
+
+#include "absl/strings/str_join.h"
+#include "absl/strings/str_split.h"
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "tensorflow/lite/c/common.h"
+#include "tensorflow/lite/kernels/internal/compatibility.h"
+#include "tensorflow/lite/tools/logging.h"
+#include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
+
+namespace tflite {
+namespace metadata {
+namespace {
+
+// Members that are added to the metadata schema after the initial version
+// of 1.0.0.
+enum class SchemaMembers {
+  kAssociatedFileTypeVocabulary = 0,
+  kSubGraphMetadataInputProcessUnits = 1,
+  kSubGraphMetadataOutputProcessUnits = 2,
+  kProcessUnitOptionsBertTokenizerOptions = 3,
+  kProcessUnitOptionsSentencePieceTokenizerOptions = 4,
+  kSubGraphMetadataInputTensorGroups = 5,
+  kSubGraphMetadataOutputTensorGroups = 6,
+  kProcessUnitOptionsRegexTokenizerOptions = 7,
+};
+
+// Helper class to compare semantic versions in terms of three integers, major,
+// minor, and patch.
+class Version {
+ public:
+  explicit Version(int major, int minor = 0, int patch = 0)
+      : version_({major, minor, patch}) {}
+
+  explicit Version(const std::string& version) {
+    const std::vector<std::string> vec = absl::StrSplit(version, '.');
+    // The version string should always be less than four numbers.
+    TFLITE_DCHECK(vec.size() <= kElementNumber && !vec.empty());
+    version_[0] = std::stoi(vec[0]);
+    version_[1] = vec.size() > 1 ? std::stoi(vec[1]) : 0;
+    version_[2] = vec.size() > 2 ? std::stoi(vec[2]) : 0;
+  }
+
+  // Compares two semantic version numbers.
+  //
+  // Example results when comparing two versions strings:
+  //   "1.9" precedes "1.14";
+  //   "1.14" precedes "1.14.1";
+  //   "1.14" and "1.14.0" are equal.
+  //
+  // Returns the value 0 if the two versions are equal; a value less than 0 if
+  // *this precedes v; a value greater than 0 if v precedes *this.
+  int Compare(const Version& v) {
+    for (int i = 0; i < kElementNumber; ++i) {
+      if (version_[i] != v.version_[i]) {
+        return version_[i] < v.version_[i] ? -1 : 1;
+      }
+    }
+    return 0;
+  }
+
+  // Converts version_ into a version string.
+  std::string ToString() { return absl::StrJoin(version_, "."); }
+
+ private:
+  static constexpr int kElementNumber = 3;
+  std::array<int, kElementNumber> version_;
+};
+
+Version GetMemberVersion(SchemaMembers member) {
+  switch (member) {
+    case SchemaMembers::kAssociatedFileTypeVocabulary:
+      return Version(1, 0, 1);
+    case SchemaMembers::kSubGraphMetadataInputProcessUnits:
+      return Version(1, 1, 0);
+    case SchemaMembers::kSubGraphMetadataOutputProcessUnits:
+      return Version(1, 1, 0);
+    case SchemaMembers::kProcessUnitOptionsBertTokenizerOptions:
+      return Version(1, 1, 0);
+    case SchemaMembers::kProcessUnitOptionsSentencePieceTokenizerOptions:
+      return Version(1, 1, 0);
+    case SchemaMembers::kSubGraphMetadataInputTensorGroups:
+      return Version(1, 2, 0);
+    case SchemaMembers::kSubGraphMetadataOutputTensorGroups:
+      return Version(1, 2, 0);
+    case SchemaMembers::kProcessUnitOptionsRegexTokenizerOptions:
+      return Version(1, 2, 1);
+    default:
+      // Should never happen.
+      TFLITE_LOG(FATAL) << "Unsupported schema member: "
+                        << static_cast<int>(member);
+  }
+  // Should never happen.
+  return Version(0, 0, 0);
+}
+
+// Updates min_version if it precedes the new_version.
+inline void UpdateMinimumVersion(const Version& new_version,
+                                 Version* min_version) {
+  if (min_version->Compare(new_version) < 0) {
+    *min_version = new_version;
+  }
+}
+
+template <typename T>
+void UpdateMinimumVersionForTable(const T* table, Version* min_version);
+
+template <typename T>
+void UpdateMinimumVersionForArray(
+    const flatbuffers::Vector<flatbuffers::Offset<T>>* array,
+    Version* min_version) {
+  if (array == nullptr) return;
+
+  for (int i = 0; i < array->size(); ++i) {
+    UpdateMinimumVersionForTable<T>(array->Get(i), min_version);
+  }
+}
+
+template <>
+void UpdateMinimumVersionForTable<tflite::AssociatedFile>(
+    const tflite::AssociatedFile* table, Version* min_version) {
+  if (table == nullptr) return;
+
+  if (table->type() == AssociatedFileType_VOCABULARY) {
+    UpdateMinimumVersion(
+        GetMemberVersion(SchemaMembers::kAssociatedFileTypeVocabulary),
+        min_version);
+  }
+}
+
+template <>
+void UpdateMinimumVersionForTable<tflite::ProcessUnit>(
+    const tflite::ProcessUnit* table, Version* min_version) {
+  if (table == nullptr) return;
+
+  tflite::ProcessUnitOptions process_unit_type = table->options_type();
+  if (process_unit_type == ProcessUnitOptions_BertTokenizerOptions) {
+    UpdateMinimumVersion(
+        GetMemberVersion(
+            SchemaMembers::kProcessUnitOptionsBertTokenizerOptions),
+        min_version);
+  }
+  if (process_unit_type == ProcessUnitOptions_SentencePieceTokenizerOptions) {
+    UpdateMinimumVersion(
+        GetMemberVersion(
+            SchemaMembers::kProcessUnitOptionsSentencePieceTokenizerOptions),
+        min_version);
+  }
+  if (process_unit_type == ProcessUnitOptions_RegexTokenizerOptions) {
+    UpdateMinimumVersion(
+        GetMemberVersion(
+            SchemaMembers::kProcessUnitOptionsRegexTokenizerOptions),
+        min_version);
+  }
+}
+
+template <>
+void UpdateMinimumVersionForTable<tflite::TensorMetadata>(
+    const tflite::TensorMetadata* table, Version* min_version) {
+  if (table == nullptr) return;
+
+  // Checks the associated_files field.
+  UpdateMinimumVersionForArray<tflite::AssociatedFile>(
+      table->associated_files(), min_version);
+
+  // Checks the process_units field.
+  UpdateMinimumVersionForArray<tflite::ProcessUnit>(table->process_units(),
+                                                    min_version);
+}
+
+template <>
+void UpdateMinimumVersionForTable<tflite::SubGraphMetadata>(
+    const tflite::SubGraphMetadata* table, Version* min_version) {
+  if (table == nullptr) return;
+
+  // Checks in the input/output metadata arrays.
+  UpdateMinimumVersionForArray<tflite::TensorMetadata>(
+      table->input_tensor_metadata(), min_version);
+  UpdateMinimumVersionForArray<tflite::TensorMetadata>(
+      table->output_tensor_metadata(), min_version);
+
+  // Checks the associated_files field.
+  UpdateMinimumVersionForArray<tflite::AssociatedFile>(
+      table->associated_files(), min_version);
+
+  // Checks for the input_process_units field.
+  if (table->input_process_units() != nullptr) {
+    UpdateMinimumVersion(
+        GetMemberVersion(SchemaMembers::kSubGraphMetadataInputProcessUnits),
+        min_version);
+    UpdateMinimumVersionForArray<tflite::ProcessUnit>(
+        table->input_process_units(), min_version);
+  }
+
+  // Checks for the output_process_units field.
+  if (table->output_process_units() != nullptr) {
+    UpdateMinimumVersion(
+        GetMemberVersion(SchemaMembers::kSubGraphMetadataOutputProcessUnits),
+        min_version);
+    UpdateMinimumVersionForArray<tflite::ProcessUnit>(
+        table->output_process_units(), min_version);
+  }
+
+  // Checks for the input_tensor_groups field.
+  if (table->input_tensor_groups() != nullptr) {
+    UpdateMinimumVersion(
+        GetMemberVersion(SchemaMembers::kSubGraphMetadataInputTensorGroups),
+        min_version);
+  }
+
+  // Checks for the output_tensor_groups field.
+  if (table->output_tensor_groups() != nullptr) {
+    UpdateMinimumVersion(
+        GetMemberVersion(SchemaMembers::kSubGraphMetadataOutputTensorGroups),
+        min_version);
+  }
+}
+
+template <>
+void UpdateMinimumVersionForTable<tflite::ModelMetadata>(
+    const tflite::ModelMetadata* table, Version* min_version) {
+  if (table == nullptr) {
+    // Should never happen, because VerifyModelMetadataBuffer has verified it.
+    TFLITE_LOG(FATAL) << "The ModelMetadata object is null.";
+    return;
+  }
+
+  // Checks the subgraph_metadata field.
+  if (table->subgraph_metadata() != nullptr) {
+    for (int i = 0; i < table->subgraph_metadata()->size(); ++i) {
+      UpdateMinimumVersionForTable<tflite::SubGraphMetadata>(
+          table->subgraph_metadata()->Get(i), min_version);
+    }
+  }
+
+  // Checks the associated_files field.
+  UpdateMinimumVersionForArray<tflite::AssociatedFile>(
+      table->associated_files(), min_version);
+}
+
+}  // namespace
+
+TfLiteStatus GetMinimumMetadataParserVersion(const uint8_t* buffer_data,
+                                             size_t buffer_size,
+                                             std::string* min_version_str) {
+  flatbuffers::Verifier verifier =
+      flatbuffers::Verifier(buffer_data, buffer_size);
+  if (!tflite::VerifyModelMetadataBuffer(verifier)) {
+    TFLITE_LOG(ERROR) << "The model metadata is not a valid FlatBuffer buffer.";
+    return kTfLiteError;
+  }
+
+  static constexpr char kDefaultVersion[] = "1.0.0";
+  Version min_version = Version(kDefaultVersion);
+
+  // Checks if any member declared after 1.0.0 (such as those in
+  // SchemaMembers) exists, and updates min_version accordingly. The minimum
+  // metadata parser version will be the largest version number of all fields
+  // that has been added to a metadata flatbuffer
+  const tflite::ModelMetadata* model_metadata = GetModelMetadata(buffer_data);
+
+  // All tables in the metadata schema should have their dedicated
+  // UpdateMinimumVersionForTable<Foo>() methods, respectively. We'll gradually
+  // add these methods when new fields show up in later schema versions.
+  //
+  // UpdateMinimumVersionForTable<Foo>() takes a const pointer of Foo. The
+  // pointer can be a nullptr if Foo is not populated into the corresponding
+  // table of the Flatbuffer object. In this case,
+  // UpdateMinimumVersionFor<Foo>() will be skipped. An exception is
+  // UpdateMinimumVersionForModelMetadata(), where ModelMetadata is the root
+  // table, and it won't be null.
+  UpdateMinimumVersionForTable<tflite::ModelMetadata>(model_metadata,
+                                                      &min_version);
+
+  *min_version_str = min_version.ToString();
+  return kTfLiteOk;
+}
+
+}  // namespace metadata
+}  // namespace tflite
diff --git a/tensorflow_lite_support/metadata/cc/metadata_version.h b/tensorflow_lite_support/metadata/cc/metadata_version.h
new file mode 100644
index 0000000..6332aae
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/metadata_version.h
@@ -0,0 +1,38 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+#ifndef TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_VERSION_H_
+#define TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_VERSION_H_
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include <string>
+
+#include "tensorflow/lite/c/common.h"
+
+namespace tflite {
+namespace metadata {
+
+// Gets the minimum metadata parser version that can fully understand all fields
+// in a given metadata flatbuffer. TFLite Metadata follows Semantic Versioning
+// 2.0. Each release version has the form MAJOR.MINOR.PATCH.
+TfLiteStatus GetMinimumMetadataParserVersion(const uint8_t* buffer_data,
+                                             size_t buffer_size,
+                                             std::string* min_version);
+
+}  // namespace metadata
+}  // namespace tflite
+
+#endif  // TENSORFLOW_LITE_SUPPORT_METADATA_CC_METADATA_VERSION_H_
diff --git a/tensorflow_lite_support/metadata/cc/python/BUILD b/tensorflow_lite_support/metadata/cc/python/BUILD
new file mode 100644
index 0000000..34e9a4f
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/python/BUILD
@@ -0,0 +1,22 @@
+load("@org_tensorflow//tensorflow:tensorflow.bzl", "pybind_extension")
+
+package(
+    default_visibility = [
+        "//tensorflow_lite_support/metadata:__subpackages__",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+pybind_extension(
+    name = "_pywrap_metadata_version",
+    srcs = [
+        "metadata_version.cc",
+    ],
+    features = ["-use_header_modules"],
+    module_name = "_pywrap_metadata_version",
+    deps = [
+        "//tensorflow_lite_support/metadata/cc:metadata_version",
+        "@org_tensorflow//tensorflow/lite/c:common",
+        "@pybind11",
+    ],
+)
diff --git a/tensorflow_lite_support/metadata/cc/python/metadata_version.cc b/tensorflow_lite_support/metadata/cc/python/metadata_version.cc
new file mode 100644
index 0000000..db3a29e
--- /dev/null
+++ b/tensorflow_lite_support/metadata/cc/python/metadata_version.cc
@@ -0,0 +1,55 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "tensorflow_lite_support/metadata/cc/metadata_version.h"
+
+#include "pybind11/pybind11.h"
+#include "tensorflow/lite/c/common.h"
+
+namespace tflite {
+namespace metadata {
+
+PYBIND11_MODULE(_pywrap_metadata_version, m) {
+  m.doc() = R"pbdoc(
+    _pywrap_metadata_version
+    A module that returns the minimum metadata parser version of a given
+    metadata flatbuffer.
+  )pbdoc";
+
+  // Using pybind11 type conversions to convert between Python and native
+  // C++ types. There are other options to provide access to native Python types
+  // in C++ and vice versa. See the pybind 11 instrcution [1] for more details.
+  // Type converstions is recommended by pybind11, though the main downside
+  // is that a copy of the data must be made on every Python to C++ transition:
+  // this is needed since the C++ and Python versions of the same type generally
+  // won’t have the same memory layout.
+  //
+  // [1]: https://pybind11.readthedocs.io/en/stable/advanced/cast/index.html
+  m.def("GetMinimumMetadataParserVersion",
+        [](const std::string& buffer_data) -> std::string {
+          std::string min_version;
+          if (GetMinimumMetadataParserVersion(
+                  reinterpret_cast<const uint8_t*>(buffer_data.c_str()),
+                  buffer_data.length(), &min_version) != kTfLiteOk) {
+            pybind11::value_error(
+                "Error occurred when getting the minimum metadata parser "
+                "version of the metadata flatbuffer.");
+          }
+          return min_version;
+        });
+}
+
+}  // namespace metadata
+}  // namespace tflite
diff --git a/tensorflow_lite_support/metadata/flatbuffers_lib/BUILD b/tensorflow_lite_support/metadata/flatbuffers_lib/BUILD
new file mode 100644
index 0000000..d4171bf
--- /dev/null
+++ b/tensorflow_lite_support/metadata/flatbuffers_lib/BUILD
@@ -0,0 +1,22 @@
+load("@org_tensorflow//tensorflow:tensorflow.bzl", "pybind_extension")
+
+package(
+    default_visibility = [
+        "//visibility:public",
+    ],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+pybind_extension(
+    name = "_pywrap_flatbuffers",
+    srcs = [
+        "flatbuffers_lib.cc",
+    ],
+    features = ["-use_header_modules"],
+    module_name = "_pywrap_flatbuffers",
+    deps = [
+        "@flatbuffers",
+        "@local_config_python//:python_headers",
+        "@pybind11",
+    ],
+)
diff --git a/tensorflow_lite_support/metadata/flatbuffers_lib/flatbuffers_lib.cc b/tensorflow_lite_support/metadata/flatbuffers_lib/flatbuffers_lib.cc
new file mode 100644
index 0000000..6185722
--- /dev/null
+++ b/tensorflow_lite_support/metadata/flatbuffers_lib/flatbuffers_lib.cc
@@ -0,0 +1,59 @@
+/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+#include "flatbuffers/flatbuffers.h"  // from @flatbuffers
+#include "flatbuffers/idl.h"  // from @flatbuffers
+#include "pybind11/pybind11.h"
+#include "pybind11/pytypes.h"
+#include "pybind11/stl.h"
+
+namespace tflite {
+namespace support {
+
+PYBIND11_MODULE(_pywrap_flatbuffers, m) {
+  pybind11::class_<flatbuffers::IDLOptions>(m, "IDLOptions")
+      .def(pybind11::init<>())
+      .def_readwrite("strict_json", &flatbuffers::IDLOptions::strict_json);
+  pybind11::class_<flatbuffers::Parser>(m, "Parser")
+      .def(pybind11::init<const flatbuffers::IDLOptions&>())
+      .def("parse",
+           [](flatbuffers::Parser* self, const std::string& source) {
+             return self->Parse(source.c_str());
+           })
+      .def_readonly("builder", &flatbuffers::Parser::builder_)
+      .def_readonly("error", &flatbuffers::Parser::error_);
+  pybind11::class_<flatbuffers::FlatBufferBuilder>(m, "FlatBufferBuilder")
+      .def("clear", &flatbuffers::FlatBufferBuilder::Clear)
+      .def("push_flat_buffer", [](flatbuffers::FlatBufferBuilder* self,
+                                  const std::string& contents) {
+        self->PushFlatBuffer(reinterpret_cast<const uint8_t*>(contents.c_str()),
+                             contents.length());
+      });
+  m.def("generate_text_file", &flatbuffers::GenerateTextFile);
+  m.def(
+      "generate_text",
+      [](const flatbuffers::Parser& parser,
+         const std::string& buffer) -> std::string {
+        std::string text;
+        if (!flatbuffers::GenerateText(
+                parser, reinterpret_cast<const void*>(buffer.c_str()), &text)) {
+          return "";
+        }
+        return text;
+      });
+}
+
+}  // namespace support
+}  // namespace tflite
diff --git a/tensorflow_lite_support/metadata/metadata_schema.fbs b/tensorflow_lite_support/metadata/metadata_schema.fbs
new file mode 100644
index 0000000..8faae0a
--- /dev/null
+++ b/tensorflow_lite_support/metadata/metadata_schema.fbs
@@ -0,0 +1,686 @@
+// Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+namespace tflite;
+
+// TFLite metadata contains both human readable and machine readable information
+// about what the model does and how to use the model. It can be used as a
+// README file, which elaborates the details of the model, each input/ouput
+// tensor, and each associated file.
+//
+// An important use case of TFLite metadata is the TFLite codegen tool, which
+// automatically generates the model interface based on the properties of the
+// model and the tensors. The model interface provides high-level APIs to
+// interact with the model, such as preprocessing the input data and running
+// inferences.
+//
+// Entries marked with "<Codegen usage>" are used in TFLite codegen tool to
+// generate the model interface. It is recommended to fill in at least those
+// enties to boost the codegen performance.
+
+// The Metadata schema is versioned by the Semantic versioning number, such as
+// MAJOR.MINOR.PATCH. It tracks the schema changes according to the rules below:
+//  * Bump up the MAJOR number when making potentially backwards incompatible
+//    changes. It must be incremented if the new changes break the backwards
+//    compatibility. It may also include minor and patch level changes as
+//    needed. The true backwards compatibility is indicated by the file
+//    identifier.
+//  * Bump up the MINOR number when making backwards compatible updates for
+//    major features, such as supporting new content types or adding new
+//    processing units.
+//  * Bump up the PATCH number when making small backwards compatible changes,
+//    such as adding a new fields or deprecating certain fields (not deleting
+//    them).
+//
+// ModelMetadata.min_parser_version indicates the minimum necessary metadata
+// parser version to fully understand all fields in a given metadata flatbuffer.
+//
+// New fields and types will have associated comments with the schema version
+// for which they were added.
+//
+// LINT.IfChange
+// Schema Semantic version: 1.2.1
+// LINT.ThenChange(//tensorflow_lite_support/\
+//     metadata/java/src/java/org/tensorflow/lite/support/metadata/\
+//     MetadataParser.java)
+
+// This indicates the flatbuffer compatibility. The number will bump up when a
+// break change is applied to the schema, such as removing fields or adding new
+// fields to the middle of a table.
+file_identifier "M001";
+
+// History:
+// 1.0.1 - Added VOCABULARY type to AssociatedFileType.
+// 1.1.0 - Added BertTokenizerOptions to ProcessUnitOptions.
+//         Added SentencePieceTokenizerOptions to ProcessUnitOptions.
+//         Added input_process_units to SubGraphMetadata.
+//         Added output_process_units to SubGraphMetadata.
+// 1.2.0 - Added input_tensor_group to SubGraphMetadata.
+//         Added output_tensor_group to SubGraphMetadata.
+// 1.2.1 - Added RegexTokenizerOptions to ProcessUnitOptions.
+
+// File extension of any written files.
+file_extension "tflitemeta";
+
+// LINT.IfChange
+enum AssociatedFileType : byte {
+  UNKNOWN = 0,
+
+  // Files such as readme.txt.
+  DESCRIPTIONS = 1,
+
+  // Contains labels that annotate certain axis of the tensor. For example,
+  // the label file in image classification. Those labels annotate the
+  // the output tensor, such that each value in the output tensor is the
+  // probability of that corresponding category specified by the label.
+  //
+  // <Codegen usage>:
+  // If an output tensor has an associated file as TENSOR_AXIS_LABELS, return
+  // the output as a mapping between the labels and probability in the model
+  // interface.
+  // If multiple files of the same type are present, the first one is used by
+  // default; additional ones are to be distinguished from one another by their
+  // specified locale.
+  TENSOR_AXIS_LABELS = 2,
+
+  // Contains labels that tensor values correspond to. For example, in
+  // the object detection model, one of the output tensors is the detected
+  // classes. And each value in the tensor refers to the index of label in the
+  // category label file.
+  //
+  // <Codegen usage>:
+  // If an output tensor has an associated file as TENSOR_VALUE_LABELS, convert
+  // the tensor values into labels, and return a list of string as the output.
+  // If multiple files of the same type are present, the first one is used by
+  // default; additional ones are to be distinguished from one another by their
+  // specified locale.
+  TENSOR_VALUE_LABELS = 3,
+
+  // Contains sigmoid-based score calibration parameters, formatted as CSV.
+  // Lines contain for each index of an output tensor the scale, slope, offset
+  // and (optional) min_score parameters to be used for sigmoid fitting (in this
+  // order and in `strtof`-compatible [1] format).
+  // A line may be left empty to default calibrated scores for this index to
+  // default_score.
+  // In summary, each line should thus contain 0, 3 or 4 comma-separated values.
+  //
+  // See documentation for ScoreCalibrationOptions for details.
+  //
+  // [1]: https://en.cppreference.com/w/c/string/byte/strtof
+  TENSOR_AXIS_SCORE_CALIBRATION = 4,
+
+  // Contains a list of unique words (characters separated by "\n" or in lines)
+  // that help to convert natural language words to embedding vectors.
+  // Added in: 1.0.1
+  VOCABULARY = 5,
+}
+
+table AssociatedFile {
+  // Name of this file. Need to be exact the same as the name of the actual file
+  // packed into the TFLite model as a zip file.
+  //
+  // <Codegen usage>:
+  // Locates to the actual file in the TFLite model.
+  name:string;
+
+  // A description of what the file is.
+  description:string;
+
+  // Type of the associated file. There may be special pre/post processing for
+  // some types. For example in image classification, a label file of the output
+  // will be used to convert object index into string.
+  //
+  // <Codegen usage>:
+  // Determines how to process the corresponding tensor.
+  type:AssociatedFileType;
+
+  // An optional locale for this associated file (if applicable). It is
+  // recommended to use an ISO 639-1 letter code (e.g. "en" for English),
+  // optionally completed by a two letter region code (e.g. "en-US" for US
+  // English and "en-CA" for Canadian English).
+  // Leverage this in order to specify e.g multiple label files translated in
+  // different languages.
+  locale:string;
+}
+
+// The basic content type for all tensors.
+//
+// <Codegen usage>:
+// Input feature tensors:
+// 1. Generates the method to load data from a TensorBuffer.
+// 2. Creates the preprocessing logic. The default processing pipeline is:
+// [NormalizeOp, QuantizeOp].
+// Output feature tensors:
+// 1. Generates the method to return the output data to a TensorBuffer.
+// 2. Creates the post-processing logic. The default processing pipeline is:
+// [DeQuantizeOp].
+table FeatureProperties {
+}
+
+// The type of color space of an image.
+enum ColorSpaceType : byte {
+  UNKNOWN = 0,
+  RGB = 1,
+  GRAYSCALE = 2,
+}
+
+table ImageSize {
+  width:uint;
+  height:uint;
+}
+
+// The properties for image tensors.
+//
+// <Codegen usage>:
+// Input image tensors:
+// 1. Generates the method to load an image from a TensorImage.
+// 2. Creates the preprocessing logic. The default processing pipeline is:
+// [ResizeOp, NormalizeOp, QuantizeOp].
+// Output image tensors:
+// 1. Generates the method to return the output data to a TensorImage.
+// 2. Creates the post-processing logic. The default processing pipeline is:
+// [DeQuantizeOp].
+table ImageProperties {
+  // The color space of the image.
+  //
+  // <Codegen usage>:
+  // Determines how to convert the color space of a given image from users.
+  color_space:ColorSpaceType;
+
+  // Indicates the default value of image width and height if the tensor shape
+  // is dynamic. For fixed-size tensor, this size will be consistent with the
+  // expected size.
+  default_size:ImageSize;
+}
+
+// The properties for tensors representing bounding boxes.
+//
+// <Codegen usage>:
+// Input image tensors: NA.
+// Output image tensors: parses the values into a data stucture that represents
+// bounding boxes. For example, in the generated wrapper for Android, it returns
+// the output as android.graphics.Rect objects.
+enum BoundingBoxType : byte {
+  UNKNOWN = 0,
+  // Represents the bounding box by using the combination of boundaries,
+  // {left, top, right, bottom}.
+  // The default order is {left, top, right, bottom}. Other orders can be
+  // indicated by BoundingBoxProperties.index.
+  BOUNDARIES = 1,
+
+  // Represents the bounding box by using the upper_left corner, width and
+  // height.
+  // The default order is {upper_left_x, upper_left_y, width, height}. Other
+  // orders can be indicated by BoundingBoxProperties.index.
+  UPPER_LEFT = 2,
+
+  // Represents the bounding box by using the center of the box, width and
+  // height. The default order is {center_x, center_y, width, height}. Other
+  // orders can be indicated by BoundingBoxProperties.index.
+  CENTER = 3,
+
+}
+
+enum CoordinateType : byte {
+  // The coordinates are float values from 0 to 1.
+  RATIO = 0,
+  // The coordinates are integers.
+  PIXEL = 1,
+}
+
+table BoundingBoxProperties {
+  // Denotes the order of the elements defined in each bounding box type. An
+  // empty index array represent the default order of each bounding box type.
+  // For example, to denote the default order of BOUNDARIES, {left, top, right,
+  // bottom}, the index should be {0, 1, 2, 3}. To denote the order {left,
+  // right, top, bottom}, the order should be {0, 2, 1, 3}.
+  //
+  // The index array can be applied to all bounding box types to adjust the
+  // order of their corresponding underlying elements.
+  //
+  // <Codegen usage>:
+  // Indicates how to parse the bounding box values.
+  index:[uint];
+
+  // <Codegen usage>:
+  // Indicates how to parse the bounding box values.
+  type:BoundingBoxType;
+
+  // <Codegen usage>:
+  // Indicates how to convert the bounding box back to the original image in
+  // pixels.
+  coordinate_type:CoordinateType;
+}
+
+union ContentProperties {
+  FeatureProperties,
+  ImageProperties,
+  BoundingBoxProperties,
+}
+
+table ValueRange {
+  min:int;
+  max:int;
+}
+
+table Content {
+  // The properties that the content may have, indicating the type of the
+  // Content.
+  //
+  // <Codegen usage>:
+  // Indicates how to process the tensor.
+  content_properties:ContentProperties;
+
+  // The range of dimensions that the content corresponds to. A NULL
+  // "range" indicates that the content uses up all dimensions,
+  // except the batch axis if applied.
+  //
+  // Here are all the possible situations of how a tensor is composed.
+  // Case 1: The tensor is a single object, such as an image.
+  // For example, the input of an image classifier
+  // (https://www.tensorflow.org/lite/models/image_classification/overview),
+  // a tensor of shape [1, 224, 224, 3]. Dimensions 1 to 3 correspond to the
+  // image. Since dimension 0 is a batch axis, which can be ignored,
+  // "range" can be left as NULL.
+  //
+  // Case 2: The tensor contains multiple instances of the same object.
+  // For example, the output tensor of detected bounding boxes of an object
+  // detection model
+  // (https://www.tensorflow.org/lite/models/object_detection/overview).
+  // The tensor shape is [1, 10, 4]. Here is the what the three dimensions
+  // represent for:
+  // dimension 0: the batch axis.
+  // dimension 1: the 10 objects detected with the highest confidence.
+  // dimension 2: the bounding boxes of the 10 detected objects.
+  // The tensor is essentially 10 bounding boxes. In this case,
+  // "range" should be {min=2; max=2;}.
+  //
+  // The output tensor of scores of the above object detection model has shape
+  // [1, 10], where
+  // dimension 0: the batch axis;
+  // dimension 1: the scores of the 10 detected objects.
+  // Set "range" to the number of dimensions which is {min=2; max=2;} to denote
+  // that every element in the tensor is an individual content object, i.e. a
+  // score in this example.
+  //
+  // Another example is the pose estimation model
+  // (https://www.tensorflow.org/lite/models/pose_estimation/overview).
+  // The output tensor of heatmaps is in the shape of [1, 9, 9, 17].
+  // Here is the what the four dimensions represent for:
+  // dimension 0: the batch axis.
+  // dimension 1/2: the heatmap image.
+  // dimension 3: 17 body parts of a person.
+  // Even though the last axis is body part, the real content of this tensor is
+  // the heatmap. "range" should be [min=1; max=2].
+  //
+  // Case 3: The tensor contains multiple different objects. (Not supported by
+  // Content at this point).
+  // Sometimes a tensor may contain multiple different objects, thus different
+  // contents. It is very common for regression models. For example, a model
+  // to predict the fuel efficiency
+  // (https://www.tensorflow.org/tutorials/keras/regression).
+  // The input tensor has shape [1, 9], consisting of 9 features, such as
+  // "Cylinders", "Displacement", "Weight", etc. In this case, dimension 1
+  // contains 9 different contents. However, since these sub-dimension objects
+  // barely need to be specifically processed, their contents are not recorded
+  // in the metadata. Through, the name of each dimension can be set through
+  // TensorMetadata.dimension_names.
+  //
+  // Note that if it is not case 3, a tensor can only have one content type.
+  //
+  // <Codegen usage>:
+  // Case 1: return a processed single object of certain content type.
+  // Case 2: return a list of processed objects of certain content type. The
+  // generated model interface have API to random access those objects from
+  // the output.
+  range:ValueRange;
+}
+
+// Parameters that are used when normalizing the tensor.
+table NormalizationOptions{
+  // mean and std are normalization parameters. Tensor values are normalized
+  // on a per-channel basis, by the formula
+  //   (x - mean) / std.
+  // If there is only one value in mean or std, we'll propogate the value to
+  // all channels.
+  //
+  // Quantized models share the same normalization parameters as their
+  // corresponding float models. For example, an image input tensor may have
+  // the normalization parameter of
+  //   mean = 127.5f and std = 127.5f.
+  // The image value will be normalized from [0, 255] to [-1, 1].
+  // Then, for quantized models, the image data should be further quantized
+  // according to the quantization parameters. In the case of uint8, the image
+  // data will be scaled back to [0, 255], while for int8, the image data will
+  // be scaled to [-128, 127].
+  //
+  // Both the normalization parameters and quantization parameters can be
+  // retrieved through the metadata extractor library.
+  // TODO(b/156644598): add link for the metadata extractor library.
+
+  // Per-channel mean of the possible values used in normalization.
+  //
+  // <Codegen usage>:
+  // Apply normalization to input tensors accordingly.
+  mean:[float];
+
+  // Per-channel standard dev. of the possible values used in normalization.
+  //
+  // <Codegen usage>:
+  // Apply normalization to input tensors accordingly.
+  std:[float];
+}
+
+// The different possible score transforms to apply to uncalibrated scores
+// before applying score calibration.
+enum ScoreTransformationType : byte {
+  // Identity function: g(x) = x.
+  IDENTITY = 0,
+  // Log function: g(x) = log(x).
+  LOG = 1,
+  // Inverse logistic function: g(x) = log(x) - log(1-x).
+  INVERSE_LOGISTIC = 2,
+}
+
+// Options to perform score calibration on an output tensor through sigmoid
+// functions. One of the main purposes of score calibration is to make scores
+// across classes comparable, so that a common threshold can be used for all
+// output classes. This is meant for models producing class predictions as
+// output, e.g. image classification or detection models.
+//
+// For each index in the output tensor, this applies:
+// * `f(x) = scale / (1 + e^-(slope*g(x)+offset))` if `x > min_score` or if no
+//   `min_score` has been specified,
+// * `f(x) = default_score` otherwise or if no scale, slope and offset have been
+//   specified.
+// Where:
+// * scale, slope, offset and (optional) min_score are index-specific parameters
+// * g(x) is an index-independent transform among those defined in
+//   ScoreTransformationType
+// * default_score is an index-independent parameter.
+// An AssociatedFile with type TANSOR_AXIS_SCORE_CALIBRATION specifying the
+// index-specific parameters must be associated with the corresponding
+// TensorMetadata for score calibration be applied.
+table ScoreCalibrationOptions {
+  // The function to use for transforming the uncalibrated score before
+  // applying score calibration.
+  score_transformation:ScoreTransformationType;
+
+  // The default calibrated score to apply if the uncalibrated score is
+  // below min_score or if no parameters were specified for a given index.
+  default_score:float;
+}
+
+// Performs thresholding on output tensor values, in order to filter out
+// low-confidence results.
+table ScoreThresholdingOptions {
+  // The recommended global threshold below which results are considered
+  // low-confidence and should be filtered out.
+  global_score_threshold:float;
+}
+
+// Performs Bert tokenization as in tf.text.BertTokenizer
+// (https://github.com/tensorflow/text/blob/3599f6fcd2b780a2dc413b90fb9315464f10b314/docs/api_docs/python/text/BertTokenizer.md)
+// Added in: 1.1.0
+table BertTokenizerOptions {
+  // The vocabulary files used in the BertTokenizer.
+  vocab_file:[AssociatedFile];
+}
+
+// Performs SentencePiece tokenization as in tf.text.SentencepieceTokenizer
+// (https://github.com/tensorflow/text/blob/3599f6fcd2b780a2dc413b90fb9315464f10b314/docs/api_docs/python/text/SentencepieceTokenizer.md).
+// Added in: 1.1.0
+table SentencePieceTokenizerOptions {
+  // The SentencePiece model files used in the SentencePieceTokenizer.
+  sentencePiece_model:[AssociatedFile];
+
+  // The optional vocabulary model files used in the SentencePieceTokenizer.
+  vocab_file:[AssociatedFile];
+}
+
+// Splits strings by the occurrences of delim_regex_pattern and converts the
+// tokens into ids. For example, given
+//   delim_regex_pattern: "\W+",
+//   string: "Words, words, words.",
+// the tokens after split are: "Words", "words", "words", "".
+// And then the tokens can be converted into ids according to the vocab_file.
+// Added in: 1.2.1
+table RegexTokenizerOptions {
+  delim_regex_pattern:string;
+  // The vocabulary files used to convert this tokens into ids.
+  vocab_file:[AssociatedFile];
+}
+
+// Options that are used when processing the tensor.
+union ProcessUnitOptions {
+  NormalizationOptions,
+  ScoreCalibrationOptions,
+  ScoreThresholdingOptions,
+  // Added in: 1.1.0
+  BertTokenizerOptions,
+  // Added in: 1.1.0
+  SentencePieceTokenizerOptions,
+  // Added in: 1.2.1
+  RegexTokenizerOptions
+}
+
+// A process unit that is used to process the tensor out-of-graph.
+table ProcessUnit {
+  options:ProcessUnitOptions;
+}
+
+
+// Statistics to describe a tensor.
+table Stats {
+  // Max and min are not currently used in tflite.support codegen. They mainly
+  // serve as references for users to better understand the model. They can also
+  // be used to validate model pre/post processing results.
+  // If there is only one value in max or min, we'll propogate the value to
+  // all channels.
+
+  // Per-channel maximum value of the tensor.
+  max:[float];
+
+  // Per-channel minimum value of the tensor.
+  min:[float];
+}
+
+// Metadata of a group of tensors. It may contain several tensors that will be
+// grouped together in codegen. For example, the TFLite object detection model
+// example (https://www.tensorflow.org/lite/models/object_detection/overview)
+// has four outputs: classes, scores, bounding boxes, and number of detections.
+// If the four outputs are bundled together using TensorGroup (for example,
+// named as "detection result"), the codegen tool will generate the class,
+// `DetectionResult`, which contains the class, score, and bouding box. And the
+// outputs of the model will be converted to a list of `DetectionResults` and
+// the number of detection. Note that the number of detection is a single
+// number, therefore is inappropriate for the list of `DetectionResult`.
+// Added in: 1.2.0
+table TensorGroup {
+  // Name of tensor group.
+  //
+  // <codegen usage>:
+  // Name of the joint class of the tensor group.
+  name:string;
+
+  // Names of the tensors to group together, corresponding to
+  // TensorMetadata.name.
+  //
+  // <codegen usage>:
+  // Determines which tensors will be added to this group. All tensors in the
+  // group should have the same number of elements specified by Content.range.
+  tensor_names:[string];
+}
+
+// Detailed information of an input or output tensor.
+table TensorMetadata {
+  // Name of the tensor.
+  //
+  // <Codegen usage>:
+  // The name of this tensor in the generated model interface.
+  name:string;
+
+  // A description of the tensor.
+  description:string;
+
+  // A list of names of the dimensions in this tensor. The length of
+  // dimension_names need to match the number of dimensions in this tensor.
+  //
+  // <Codegen usage>:
+  // The name of each dimension in the generated model interface. See "Case 2"
+  // in the comments of Content.range.
+  dimension_names:[string];
+
+  // The content that represents this tensor.
+  //
+  // <Codegen usage>:
+  // Determines how to process this tensor. See each item in ContentProperties
+  // for the default process units that will be applied to the tensor.
+  content:Content;
+
+  // The process units that are used to process the tensor out-of-graph.
+  //
+  // <Codegen usage>:
+  // Contains the parameters of the default processing pipeline for each content
+  // type, such as the normalization parameters in all content types. See the
+  // items under ContentProperties for the details of the default processing
+  // pipeline.
+  process_units:[ProcessUnit];
+
+  // The statistics of the tensor values.
+  stats:Stats;
+
+  // A list of associated files of this tensor.
+  //
+  // <Codegen usage>:
+  // Contains processing parameters of this tensor, such as normalization.
+  associated_files:[AssociatedFile];
+}
+
+table SubGraphMetadata {
+  // Name of the subgraph.
+  //
+  // Note that, since TFLite only support one subgraph at this moment, the
+  // Codegen tool will use the name in ModelMetadata in the generated model
+  // interface.
+  name:string;
+
+  // A description explains details about what the subgraph does.
+  description:string;
+
+  // Metadata of all input tensors used in this subgraph. It matches extactly
+  // the input tensors specified by `SubGraph.inputs` in the TFLite
+  // schema.fbs file[2]. The number of `TensorMetadata` in the array should
+  // equal to the number of indices in `SubGraph.inputs`.
+  //
+  // [2]: tensorflow/lite/schema/schema.fbs
+  // <Codegen usage>:
+  // Determines how to process the inputs.
+  input_tensor_metadata:[TensorMetadata];
+
+  // Metadata of all output tensors used in this subgraph. It matches extactly
+  // the output tensors specified by `SubGraph.outputs` in the TFLite
+  // schema.fbs file[2]. The number of `TensorMetadata` in the array should
+  // equal to the number of indices in `SubGraph.outputs`.
+  //
+  // <Codegen usage>:
+  // Determines how to process the outputs.
+  output_tensor_metadata:[TensorMetadata];
+
+  // A list of associated files of this subgraph.
+  associated_files:[AssociatedFile];
+
+  // Input process units of the subgraph. Some models may have complex pre and
+  // post processing logics where the process units do not work on one tensor at
+  // a time, but in a similar way of a TFLite graph. For example, in the
+  // MobileBert model (https://www.tensorflow.org/lite/models/bert_qa/overview),
+  // the inputs are: ids / mask / segment ids;
+  // the outputs are: end logits / start logits.
+  // The preprocessing converts the query string and the context string to the
+  // model inputs, and the post-processing converts the model outputs to the
+  // answer string.
+  // Added in: 1.1.0
+  input_process_units:[ProcessUnit];
+
+  // Output process units of the subgraph.
+  // Added in: 1.1.0
+  output_process_units:[ProcessUnit];
+
+  // Metadata of all input tensor groups used in this subgraph.
+  //
+  // <codegen usage>:
+  // Bundles the corresponding elements of the underlying input tensors together
+  // into a class, and converts those individual tensors into a list of the
+  // class objects.
+  // Added in: 1.2.0
+  input_tensor_groups:[TensorGroup];
+
+  // Metadata of all output tensor groups used in this subgraph.
+  //
+  // <codegen usage>:
+  // Bundles the corresponding elements of the underlying output tensors
+  // together into a class, and converts those individual tensors into a list of
+  // the class objects.
+  // Added in: 1.2.0
+  output_tensor_groups:[TensorGroup];
+
+}
+
+table ModelMetadata {
+  // Name of the model.
+  //
+  // <Codegen usage>:
+  // The name of the model in the generated model interface.
+  name:string;
+
+  // Model description in schema.
+  description:string;
+
+  // Version of the model that specified by model creators.
+  version:string;
+
+  // Noted that, the minimum required TFLite runtime version that the model is
+  // compatible with, has already been added as a metadata entry in tflite
+  // schema. We'll decide later if we want to move it here, and keep it with
+  // other metadata entries.
+
+  // Metadata of all the subgraphs of the model. The 0th is assumed to be the
+  // main subgraph.
+  //
+  // <Codegen usage>:
+  // Determines how to process the inputs and outputs.
+  subgraph_metadata:[SubGraphMetadata];
+
+  // The person who creates this model.
+  author:string;
+
+  // Licenses that may apply to this model.
+  license:string;
+
+  // A list of associated files of this model.
+  associated_files:[AssociatedFile];
+
+  // The minimum metadata parser version that can fully understand the fields in
+  // the metadata flatbuffer. The version is effectively the largest version
+  // number among the versions of all the fields populated and the smallest
+  // compatible version indicated by the file identifier.
+  //
+  // This field is automaticaly populated by the MetadataPopulator when
+  // the metadata is populated into a TFLite model.
+  min_parser_version:string;
+}
+// LINT.ThenChange(//tensorflow_lite_support/\
+//     metadata/cc/metadata_version.cc)
+
+root_type ModelMetadata;
diff --git a/tensorflow_lite_support/opensource/opensource_only.files b/tensorflow_lite_support/opensource/opensource_only.files
new file mode 100644
index 0000000..be42642
--- /dev/null
+++ b/tensorflow_lite_support/opensource/opensource_only.files
@@ -0,0 +1,36 @@
+tensorflow_lite_support/custom_ops/kernel/sentencepiece/native.bzl
+tensorflow_lite_support/opensource/BUILD
+tensorflow_lite_support/opensource/WORKSPACE
+tensorflow_lite_support/opensource/cc_build_defs.bzl
+tensorflow_lite_support/third_party/android/BUILD
+tensorflow_lite_support/third_party/android/android.bzl.tpl
+tensorflow_lite_support/third_party/android/android_configure.BUILD.tpl
+tensorflow_lite_support/third_party/android/android_configure.bzl
+tensorflow_lite_support/third_party/com_google_absl.BUILD
+tensorflow_lite_support/third_party/darts_clone.BUILD
+tensorflow_lite_support/third_party/fft2d/BUILD
+tensorflow_lite_support/third_party/fft2d/LICENSE
+tensorflow_lite_support/third_party/fft2d/fft.h
+tensorflow_lite_support/third_party/fft2d/fft2d.BUILD
+tensorflow_lite_support/third_party/fft2d/fft2d.h
+tensorflow_lite_support/third_party/google_toolbox_for_mac.BUILD
+tensorflow_lite_support/third_party/icu.BUILD
+tensorflow_lite_support/third_party/libyuv.BUILD
+tensorflow_lite_support/third_party/libzip.BUILD
+tensorflow_lite_support/third_party/pybind11.BUILD
+tensorflow_lite_support/third_party/python_runtime/BUILD
+tensorflow_lite_support/third_party/six.BUILD
+tensorflow_lite_support/third_party/stblib.BUILD
+tensorflow_lite_support/third_party/toolchains/java/BUILD
+tensorflow_lite_support/third_party/utf.BUILD
+tensorflow_lite_support/third_party/zlib.BUILD
+tensorflow_lite_support/tools/ci_build/build_all.sh
+tensorflow_lite_support/tools/ci_build/common.sh
+tensorflow_lite_support/tools/ci_build/common_win.bat
+tensorflow_lite_support/tools/pip_package/BUILD
+tensorflow_lite_support/tools/pip_package/MANIFEST.in
+tensorflow_lite_support/tools/pip_package/README
+tensorflow_lite_support/tools/pip_package/build_pip_package.sh
+tensorflow_lite_support/tools/pip_package/setup.py
+tensorflow_lite_support/tools/pip_package/simple_console_for_windows.py
+tensorflow_lite_support/tools/pip_package/tflite_support.__init__.py
\ No newline at end of file
diff --git a/tensorflow_lite_support/tools/BUILD b/tensorflow_lite_support/tools/BUILD
new file mode 100644
index 0000000..c3525ca
--- /dev/null
+++ b/tensorflow_lite_support/tools/BUILD
@@ -0,0 +1,20 @@
+package(
+    default_visibility = ["//visibility:public"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+py_binary(
+    name = "zip_files",
+    srcs = ["zip_files.py"],
+    python_version = "PY3",
+    visibility = ["//visibility:public"],
+    deps = [
+        "@absl_py//absl:app",
+        "@absl_py//absl/flags",
+    ],
+)
+
+py_library(
+    name = "expect_flatbuffers_installed",
+    srcs = [],
+)
diff --git a/tensorflow_lite_support/tools/build_rules/expand_template.bzl b/tensorflow_lite_support/tools/build_rules/expand_template.bzl
new file mode 100644
index 0000000..717860c
--- /dev/null
+++ b/tensorflow_lite_support/tools/build_rules/expand_template.bzl
@@ -0,0 +1,50 @@
+"""Build macro for libzip."""
+
+# forked from kythe/kythe/tools/build_rules/expand_template.bzl
+def _expand_template_impl(ctx):
+    ctx.actions.expand_template(
+        template = ctx.file.template,
+        output = ctx.outputs.out,
+        substitutions = ctx.attr.substitutions,
+    )
+
+expand_template = rule(
+    attrs = {
+        "out": attr.output(mandatory = True),
+        "substitutions": attr.string_dict(mandatory = True),
+        "template": attr.label(
+            mandatory = True,
+            allow_single_file = True,
+        ),
+    },
+    output_to_genfiles = True,
+    implementation = _expand_template_impl,
+)
+
+def cmake_substitutions(vars, defines = {}):
+    """Returns a dict of template substitutions combining `vars` and `defines`.
+
+    Args:
+      vars: will be turned into a dict replacing `${key}` and `@key@` with `value`.
+      defines: will be turned into a dict replacing `#cmakedefine` with `#define {value}`
+        if present is true, otherwise `/* #undef %s /*`.
+    Returns:
+      substitutions
+    """
+    subs = {}
+    for key, value in vars.items():
+        subs["${%s}" % (key,)] = str(value) if value != None else ""
+        subs["@%s@" % (key,)] = str(value) if value != None else ""
+
+    # TODO(shahms): Better handling of #cmakedefine delimiters and line endings to
+    # avoid the prefix-substitution problem.
+    # Potentially allow value to be: True, False, None or string.
+    #   True/False => Same as current
+    #   None       => assume no suffix value, include \n in sub and replacement
+    #   string     => use string to lookup in vars and assume ${} or @@ tail?
+    for macro, present in defines.items():
+        if present:
+            subs["#cmakedefine %s" % macro] = "#define %s" % macro
+        else:
+            subs["#cmakedefine %s" % macro] = "/* #undef %s */" % macro
+    return subs
diff --git a/tensorflow_lite_support/tools/ci_build/build_all.sh b/tensorflow_lite_support/tools/ci_build/build_all.sh
new file mode 100644
index 0000000..7e98b6c
--- /dev/null
+++ b/tensorflow_lite_support/tools/ci_build/build_all.sh
@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+# External `build_all.sh`
+
+set -ex
+
+bazel build -c opt --config=monolithic \
+    //tensorflow_lite_support/java:tensorflowlite_support \
+    //tensorflow_lite_support/codegen/python:codegen \
+    //tensorflow_lite_support/metadata/java:tensorflowlite_support_metadata_lib \
+    //tensorflow_lite_support/metadata/cc:metadata_extractor \
+    //tensorflow_lite_support/custom_ops/kernel:all \
+    //tensorflow_lite_support/custom_ops/python:tflite_text_api
+
+# Build Task libraries.
+bazel build -c opt --config=monolithic \
+    --config=android_arm64 --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a \
+    //tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/core:base-task-api.aar \
+    //tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/text:task-library-text \
+    //tensorflow_lite_support/java/src/java/org/tensorflow/lite/task/vision:task-library-vision
+
+
+# Run Metadata tests.
+bazel clean --expunge
+
+bazel test --test_output=all \
+    //tensorflow_lite_support/metadata/python/tests:metadata_test \
+    //tensorflow_lite_support/metadata/python/tests/metadata_writers:all
+
diff --git a/tensorflow_lite_support/tools/ci_build/builds/pip_smoke_test.sh b/tensorflow_lite_support/tools/ci_build/builds/pip_smoke_test.sh
new file mode 100644
index 0000000..a36f97d
--- /dev/null
+++ b/tensorflow_lite_support/tools/ci_build/builds/pip_smoke_test.sh
@@ -0,0 +1,114 @@
+#!/bin/bash
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+# Pip install TensorFlow Lite Support and run basic test on the pip package.
+
+# Important: Use msys shell to run this script on Windows.
+
+set -e
+set -x
+
+function run_smoke_test() {
+  VENV_TMP_DIR="$(mktemp -d)"
+
+  if [[ "$OSTYPE" == "msys" ]]; then
+    VENV_TMP_DIR="$(cygpath -m $VENV_TMP_DIR)"
+  fi
+
+  ${PYTHON_BIN_PATH} -m virtualenv -p ${PYTHON_BIN_PATH} "${VENV_TMP_DIR}" || \
+      die "FAILED: Unable to create virtualenv"
+
+  if [[ "$OSTYPE" == "msys" ]]; then
+    source "${VENV_TMP_DIR}/Scripts/activate" || \
+        die "FAILED: Unable to activate virtualenv "
+  else
+    source "${VENV_TMP_DIR}/bin/activate" || \
+        die "FAILED: Unable to activate virtualenv "
+  fi
+
+  # install tflite-support
+  python -m pip install ${WHL_NAME} || \
+      die "pip install (forcing to reinstall tflite-support) FAILED"
+  echo "Successfully installed pip package ${WHL_NAME}"
+
+  # Download a test model
+  export TEST_MODEL="$(pwd)/test.tflite"
+  wget https://tfhub.dev/tensorflow/lite-model/mobilenet_v1_0.75_192_quantized/1/metadata/1\?lite-format\=tflite -O "$TEST_MODEL"
+  if [[ "$OSTYPE" == "msys" ]]; then
+    TEST_MODEL=$(cygpath -m $TEST_MODEL)
+  fi
+
+  test_tfls_imports
+
+  test_codegen
+
+  # Deactivate from virtualenv.
+  deactivate || source deactivate || \
+      die "FAILED: Unable to deactivate from existing virtualenv."
+
+  echo "All smoke test passes!"
+}
+
+function test_tfls_imports() {
+  TMP_DIR=$(mktemp -d)
+  pushd "${TMP_DIR}"
+
+  # test for basic import and metadata display.
+  RET_VAL=$(python -c "from tflite_support import metadata; \
+md = metadata.MetadataDisplayer.with_model_file(\"$TEST_MODEL\"); \
+print(md.get_metadata_json())")
+
+  # just check if the model name is there.
+  if ! [[ ${RET_VAL} == *"MobileNetV1 image classifier (quantized)"* ]]; then
+    echo "Unexpected return value: ${RET_VAL}"
+    echo "PIP smoke test on virtualenv FAILED, do not upload ${WHL_NAME}."
+    return 1
+  fi
+
+  RESULT=$?
+
+  popd
+  return $RESULT
+}
+
+function test_codegen() {
+  TMP_DIR=$(mktemp -d)
+  pushd "${TMP_DIR}"
+
+  # test for basic import and metadata display.
+  tflite_codegen --model ${TEST_MODEL} --destination tmp
+  RESULT=$?
+
+  # just check if the model name is there.
+  if [[ ${RESULT} -ne 0 ]]; then
+    echo "Unexpected return value: ${RESULT}"
+    echo "PIP smoke test on virtualenv FAILED, do not upload ${WHL_NAME}."
+    return 1
+  fi
+
+  popd
+  return $RESULT
+}
+
+###########################################################################
+# Main
+###########################################################################
+if [[ -z "${1}" ]]; then
+  echo "TFLite Support WHL path not given, unable to install and test."
+  return 1
+fi
+
+WHL_NAME=${1}
+run_smoke_test
diff --git a/tensorflow_lite_support/tools/ci_build/common.sh b/tensorflow_lite_support/tools/ci_build/common.sh
new file mode 100644
index 0000000..4907bb1
--- /dev/null
+++ b/tensorflow_lite_support/tools/ci_build/common.sh
@@ -0,0 +1,96 @@
+#!/usr/bin/env bash
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+# External `common.sh`
+
+# Keep in sync with tensorflow core and configure.py.
+# TODO(b/158448780): Guard bazel version with IfChangeThenChange.
+LATEST_BAZEL_VERSION=3.1.0
+
+# Run flaky functions with retries.
+# run_with_retry cmd
+function run_with_retry {
+  eval "$1"
+  # If the command fails retry again in 60 seconds.
+  if [[ $? -ne 0 ]]; then
+    sleep 60
+    eval "$1"
+  fi
+}
+
+function die() {
+  echo "$@" 1>&2 ; exit 1;
+}
+
+# A small utility to run the command and only print logs if the command fails.
+# On success, all logs are hidden.
+function readable_run {
+  # Disable debug mode to avoid printing of variables here.
+  set +x
+  result=$("$@" 2>&1) || die "$result"
+  echo "$@"
+  echo "Command completed successfully at $(date)"
+  set -x
+}
+
+# TODO(b/158448780): Guard bazel installation with IfChangeThenChange.
+function set_bazel_outdir {
+  mkdir -p /tmpfs/bazel_output
+  export TEST_TMPDIR=/tmpfs/bazel_output
+}
+
+# Downloads bazelisk to ~/bin as `bazel`.
+function install_bazelisk {
+  date
+  case "$(uname -s)" in
+    Darwin) local name=bazelisk-darwin-amd64 ;;
+    Linux)  local name=bazelisk-linux-amd64  ;;
+    *) die "Unknown OS: $(uname -s)" ;;
+  esac
+  mkdir -p "$HOME/bin"
+  wget --no-verbose -O "$HOME/bin/bazel" \
+      "https://github.com/bazelbuild/bazelisk/releases/download/v1.3.0/$name"
+  chmod u+x "$HOME/bin/bazel"
+  if [[ ! ":$PATH:" =~ :"$HOME"/bin/?: ]]; then
+    PATH="$HOME/bin:$PATH"
+  fi
+  set_bazel_outdir
+  which bazel
+  bazel version
+  date
+}
+
+# Install the given bazel version on linux
+function update_bazel_linux {
+  if [[ -z "$1" ]]; then
+    BAZEL_VERSION=${LATEST_BAZEL_VERSION}
+  else
+    BAZEL_VERSION=$1
+  fi
+  rm -rf ~/bazel
+  mkdir ~/bazel
+
+  pushd ~/bazel
+  readable_run wget https://github.com/bazelbuild/bazel/releases/download/"${BAZEL_VERSION}"/bazel-"${BAZEL_VERSION}"-installer-linux-x86_64.sh
+  chmod +x bazel-*.sh
+  ./bazel-"${BAZEL_VERSION}"-installer-linux-x86_64.sh --user
+  rm bazel-"${BAZEL_VERSION}"-installer-linux-x86_64.sh
+  popd
+
+  PATH="/home/kbuilder/bin:$PATH"
+  set_bazel_outdir
+  which bazel
+  bazel version
+}
diff --git a/tensorflow_lite_support/tools/ci_build/common_win.bat b/tensorflow_lite_support/tools/ci_build/common_win.bat
new file mode 100644
index 0000000..35f39a7
--- /dev/null
+++ b/tensorflow_lite_support/tools/ci_build/common_win.bat
@@ -0,0 +1,29 @@
+:: Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+::
+:: Licensed under the Apache License, Version 2.0 (the "License");
+:: you may not use this file except in compliance with the License.
+:: You may obtain a copy of the License at
+::
+::     http://www.apache.org/licenses/LICENSE-2.0
+::
+:: Unless required by applicable law or agreed to in writing, software
+:: distributed under the License is distributed on an "AS IS" BASIS,
+:: WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+:: See the License for the specific language governing permissions and
+:: limitations under the License.
+:: =============================================================================
+
+:: This script is shamefully borrowed from:
+:: //third_party/tensorflow/tools/ci_build/release/common_win.bat.oss
+
+echo on
+
+@REM
+@REM Setup Bazel
+@REM
+:: Download Bazel from github and make sure its found in PATH.
+SET BAZEL_VERSION=3.1.0
+md C:\tools\bazel\
+wget -q https://github.com/bazelbuild/bazel/releases/download/%BAZEL_VERSION%/bazel-%BAZEL_VERSION%-windows-x86_64.exe -O C:/tools/bazel/bazel.exe
+SET PATH=C:\tools\bazel;%PATH%
+bazel version
diff --git a/tensorflow_lite_support/tools/ci_build/update_version.py b/tensorflow_lite_support/tools/ci_build/update_version.py
new file mode 100644
index 0000000..86fa588
--- /dev/null
+++ b/tensorflow_lite_support/tools/ci_build/update_version.py
@@ -0,0 +1,120 @@
+# lint as: python3
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License..
+# ==============================================================================
+"""Update version code in the repo.
+
+We use a python script rather than GNU tools to avoid cross-platform
+difficulties.
+
+The script takes 3 argument:
+ --src <path> a path pointing to the code repo.
+ --version <version> the new version code.
+ --nightly [default: false] when true, the version code will append a build
+   suffix (e.g. dev20201103)
+
+It should not run by bazel. Use it as a simple python script.
+"""
+
+import argparse
+import datetime
+import os
+import re
+
+SETUP_PY_PATH = "tensorflow_lite_support/tools/pip_package/setup.py"
+
+
+def replace_string_in_line(search, replace, filename):
+  """Replace the string in every line of the file in-place."""
+  with open(filename, "r") as f:
+    content = f.read()
+  with open(filename, "w") as f:
+    f.write(re.sub(search, replace, content))
+
+
+def get_current_version(path):
+  """Get the current version code from setup.py."""
+  for line in open(os.path.join(path, SETUP_PY_PATH)):
+    match = re.search("^_VERSION = '([a-z0-9\\.\\-]+)'", line)
+    if match:
+      return match.group(1)
+  print("Cannot find current version!")
+  return None
+
+
+def update_version(path, current_version, new_version):
+  """Update the version code in the codebase."""
+  # Update setup.py
+  replace_string_in_line(
+      "_VERSION = '%s'" % current_version,
+      # pep440 requires such a replacement
+      "_VERSION = '%s'" % new_version.replace("-", "."),
+      os.path.join(path, SETUP_PY_PATH))
+
+
+class CustomTimeZone(datetime.tzinfo):
+
+  def utcoffset(self, dt):
+    return -datetime.timedelta(hours=8)
+
+  def tzname(self, dt):
+    return "UTC-8"
+
+  def dst(self, dt):
+    return datetime.timedelta(0)
+
+
+def remove_build_suffix(version):
+  """Remove build suffix (if exists) from a version."""
+  if version.find("-dev") >= 0:
+    return version[:version.find("-dev")]
+  if version.find(".dev") >= 0:
+    return version[:version.find(".dev")]
+  if version.find("dev") >= 0:
+    return version[:version.find("dev")]
+  return version
+
+
+def main():
+  parser = argparse.ArgumentParser(description="Update TFLS version in repo")
+  parser.add_argument(
+      "--src",
+      help="a path pointing to the code repo",
+      required=True,
+      default="")
+  parser.add_argument("--version", help="the new SemVer code", default="")
+  parser.add_argument(
+      "--nightly",
+      help="if true, a build suffix will append to the version code. If "
+      "current version code or the <version> argument provided contains a "
+      "build suffix, the suffix will be replaced with the timestamp",
+      action="store_true")
+  args = parser.parse_args()
+
+  path = args.src
+  current_version = get_current_version(path)
+  if not current_version:
+    return
+  new_version = args.version if args.version else current_version
+  if args.nightly:
+    new_version = remove_build_suffix(new_version)
+    # Use UTC-8 rather than uncertain local time.
+    d = datetime.datetime.now(tz=CustomTimeZone())
+    new_version += "-dev" + d.strftime("%Y%m%d")
+  print("Updating version from %s to %s" % (current_version, new_version))
+  update_version(path, current_version, new_version)
+
+
+if __name__ == "__main__":
+  main()
diff --git a/tensorflow_lite_support/tools/pip_package/BUILD b/tensorflow_lite_support/tools/pip_package/BUILD
new file mode 100644
index 0000000..61df24a
--- /dev/null
+++ b/tensorflow_lite_support/tools/pip_package/BUILD
@@ -0,0 +1,55 @@
+# Description:
+#  Tools for building the TensorFlow pip package.
+
+load("@local_config_syslibs//:build_defs.bzl", "if_not_system_lib")
+
+package(default_visibility = ["//visibility:private"])
+
+COMMON_PIP_DEPS = [
+    ":licenses",
+    "MANIFEST.in",
+    "README",
+    "setup.py",
+    "//tensorflow_lite_support/codegen/python:codegen",
+    "//tensorflow_lite_support/metadata/python:metadata",
+]
+
+filegroup(
+    name = "licenses",
+    data = [
+        "//:LICENSE",
+        "@org_tensorflow//:LICENSE",
+    ] + if_not_system_lib(
+        "absl_py",
+        [
+            "@absl_py//absl:LICENSE",
+            "@absl_py//absl/logging:LICENSE",
+            "@absl_py//absl/flags:LICENSE",
+            "@absl_py//absl/testing:LICENSE",
+            "@absl_py//absl/third_party/unittest3_backport:LICENSE",
+        ],
+    ),
+)
+
+sh_binary(
+    name = "build_pip_package",
+    srcs = ["build_pip_package.sh"],
+    data = COMMON_PIP_DEPS +
+           select({
+               "@org_tensorflow//tensorflow:windows": [
+                   ":simple_console_for_windows",
+               ],
+               "//conditions:default": [
+               ],
+           }),
+)
+
+# On Windows, python binary is a zip file of runfiles tree.
+# Add everything to its data dependency for generating a runfiles tree
+# for building the pip package on Windows.
+py_binary(
+    name = "simple_console_for_windows",
+    srcs = ["simple_console_for_windows.py"],
+    data = COMMON_PIP_DEPS,
+    srcs_version = "PY2AND3",
+)
diff --git a/tensorflow_lite_support/tools/pip_package/MANIFEST.in b/tensorflow_lite_support/tools/pip_package/MANIFEST.in
new file mode 100644
index 0000000..e44f271
--- /dev/null
+++ b/tensorflow_lite_support/tools/pip_package/MANIFEST.in
@@ -0,0 +1,9 @@
+include LICENSE
+include README.md
+recursive-include * *.py
+recursive-include * *.pyd
+recursive-include * *.fbs
+recursive-include * *.so
+recursive-include * *.dylib
+recursive-include * *.dll
+recursive-include * *.lib
diff --git a/tensorflow_lite_support/tools/pip_package/README b/tensorflow_lite_support/tools/pip_package/README
new file mode 100644
index 0000000..1e1f9d5
--- /dev/null
+++ b/tensorflow_lite_support/tools/pip_package/README
@@ -0,0 +1 @@
+TensorFlow Lite Support
diff --git a/tensorflow_lite_support/tools/pip_package/build_pip_package.sh b/tensorflow_lite_support/tools/pip_package/build_pip_package.sh
new file mode 100755
index 0000000..2f962e3
--- /dev/null
+++ b/tensorflow_lite_support/tools/pip_package/build_pip_package.sh
@@ -0,0 +1,232 @@
+#!/usr/bin/env bash
+# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+set -e
+
+function is_absolute {
+  [[ "$1" = /* ]] || [[ "$1" =~ ^[a-zA-Z]:[/\\].* ]]
+}
+
+function real_path() {
+  is_absolute "$1" && echo "$1" || echo "$PWD/${1#./}"
+}
+
+function move_to_root_if_exists () {
+  arg_to_move="$1"
+  if [ -e "${arg_to_move}" ]; then
+    mv ${arg_to_move} ./
+  fi
+}
+
+function reorganize_includes() {
+  TMPDIR="${1%/}"
+}
+
+PLATFORM="$(uname -s | tr 'A-Z' 'a-z')"
+function is_windows() {
+  if [[ "${PLATFORM}" =~ (cygwin|mingw32|mingw64|msys)_nt* ]]; then
+    true
+  else
+    false
+  fi
+}
+
+function prepare_src() {
+  if [ $# -lt 1 ] ; then
+    echo "No destination dir provided"
+    exit 1
+  fi
+
+  TMPDIR="${1%/}"
+  mkdir -p "$TMPDIR"
+  EXTERNAL_INCLUDES="${TMPDIR}/tflite_support/include/external"
+
+  echo $(date) : "=== Preparing sources in dir: ${TMPDIR}"
+
+  if [ ! -d bazel-bin/tensorflow_lite_support ]; then
+    echo "Could not find bazel-bin.  Did you run from the root of the build tree?"
+    exit 1
+  fi
+
+  if is_windows; then
+    rm -rf ./bazel-bin/tensorflow_lite_support/tools/pip_package/simple_console_for_windows_unzip
+    mkdir -p ./bazel-bin/tensorflow_lite_support/tools/pip_package/simple_console_for_windows_unzip
+    echo "Unzipping simple_console_for_windows.zip to create runfiles tree..."
+    unzip -o -q ./bazel-bin/tensorflow_lite_support/tools/pip_package/simple_console_for_windows.zip -d ./bazel-bin/tensorflow_lite_support/tools/pip_package/simple_console_for_windows_unzip
+    echo "Unzip finished."
+    # runfiles structure after unzip the python binary
+    RUNFILES=bazel-bin/tensorflow_lite_support/tools/pip_package/simple_console_for_windows_unzip/runfiles/org_tensorflow_lite_support
+
+    # TODO(b/165872313): Investigate the case and remove the hack.
+    # On Windows, __init__.py are not auto genereated at directories that only
+    # contains Pybind libraries.
+    touch "$RUNFILES/tensorflow_lite_support/metadata/cc/__init__.py"
+    touch "$RUNFILES/tensorflow_lite_support/metadata/cc/python/__init__.py"
+    touch "$RUNFILES/tensorflow_lite_support/metadata/flatbuffers_lib/__init__.py"
+  else
+    RUNFILES=bazel-bin/tensorflow_lite_support/tools/pip_package/build_pip_package.runfiles/org_tensorflow_lite_support
+  fi
+
+  cp "$RUNFILES/LICENSE" "${TMPDIR}"
+  cp -R "$RUNFILES/tensorflow_lite_support" "${TMPDIR}"
+
+  reorganize_includes "${TMPDIR}"
+
+  cp tensorflow_lite_support/tools/pip_package/MANIFEST.in ${TMPDIR}
+  cp tensorflow_lite_support/tools/pip_package/README ${TMPDIR}/README.md
+  cp tensorflow_lite_support/tools/pip_package/setup.py ${TMPDIR}
+
+  # A helper entry.
+  mkdir ${TMPDIR}/tflite_support
+  cp tensorflow_lite_support/tools/pip_package/tflite_support.__init__.py ${TMPDIR}/tflite_support/__init__.py
+}
+
+function build_wheel() {
+  if [ $# -lt 2 ] ; then
+    echo "No src and dest dir provided"
+    exit 1
+  fi
+
+  TMPDIR="$1"
+  DEST="$2"
+  PKG_NAME_FLAG="$3"
+
+  # Before we leave the top-level directory, make sure we know how to
+  # call python.
+  if [[ -e tools/python_bin_path.sh ]]; then
+    source tools/python_bin_path.sh
+  fi
+
+  pushd ${TMPDIR} > /dev/null
+
+  rm -f MANIFEST
+  echo $(date) : "=== Building wheel"
+  "${PYTHON_BIN_PATH:-python}" setup.py bdist_wheel ${PKG_NAME_FLAG} >/dev/null
+  mkdir -p ${DEST}
+  cp dist/* ${DEST}
+  popd > /dev/null
+  echo $(date) : "=== Output wheel file is in: ${DEST}"
+}
+
+function usage() {
+  echo "Usage:"
+  echo "$0 [--src srcdir] [--dst dstdir] [options]"
+  echo "$0 dstdir [options]"
+  echo ""
+  echo "    --src                 prepare sources in srcdir"
+  echo "                              will use temporary dir if not specified"
+  echo ""
+  echo "    --dst                 build wheel in dstdir"
+  echo "                              if dstdir is not set do not build, only prepare sources"
+  echo ""
+  echo "  Options:"
+  echo "    --project_name <name>           set project name to <name>"
+  echo "    --version <version>             reset the pip package version to <version>"
+  echo "    --nightly_flag                  build TFLite Support nightly"
+  echo ""
+  echo "When using bazel, add the following flag: --run_under=\"cd \$PWD && \""
+  echo ""
+  exit 1
+}
+
+function main() {
+  PKG_NAME_FLAG=""
+  PROJECT_NAME=""
+  NIGHTLY_BUILD=0
+  SRCDIR=""
+  DSTDIR=""
+  CLEANSRC=1
+  VERSION=""
+  while true; do
+    if [[ "$1" == "--help" ]]; then
+      usage
+      exit 1
+    elif [[ "$1" == "--nightly_flag" ]]; then
+      NIGHTLY_BUILD=1
+    elif [[ "$1" == "--project_name" ]]; then
+      shift
+      if [[ -z "$1" ]]; then
+        break
+      fi
+      PROJECT_NAME="$1"
+    elif [[ "$1" == "--version" ]]; then
+      shift
+      if [[ -z "$1" ]]; then
+        break
+      fi
+      VERSION="$1"
+    elif [[ "$1" == "--src" ]]; then
+      shift
+      SRCDIR="$(real_path $1)"
+      CLEANSRC=0
+    elif [[ "$1" == "--dst" ]]; then
+      shift
+      DSTDIR="$(real_path $1)"
+    else
+      echo "Unrecognized flag: $1"
+      usage
+      exit 1
+    fi
+    shift
+
+    if [[ -z "$1" ]]; then
+      break
+    fi
+  done
+
+  if [[ -z "$DSTDIR" ]] && [[ -z "$SRCDIR" ]]; then
+    echo "No destination dir provided"
+    usage
+    exit 1
+  fi
+
+  if [[ -z "$SRCDIR" ]]; then
+    # make temp srcdir if none set
+    SRCDIR="$(mktemp -d -t tmp.XXXXXXXXXX)"
+  fi
+
+  if [[ -z "$DSTDIR" ]]; then
+      # only want to prepare sources
+      exit
+  fi
+
+  if [[ -n ${PROJECT_NAME} ]]; then
+    PKG_NAME_FLAG="--project_name ${PROJECT_NAME}"
+  elif [[ ${NIGHTLY_BUILD} == "1" ]]; then
+    PKG_NAME_FLAG="--project_name tflite_support_nightly"
+  fi
+
+  if [[ ${NIGHTLY_BUILD} == "1" ]]; then
+    # we use a script to update versions to avoid any tool differences on different platforms.
+    if [[ ! -z ${VERSION} ]]; then
+      python tensorflow_lite_support/tools/ci_build/update_version.py --src "." --version ${VERSION} --nightly
+    else
+      python tensorflow_lite_support/tools/ci_build/update_version.py --src "." --nightly
+    fi
+  elif [[ ! -z ${VERSION} ]]; then
+    python tensorflow_lite_support/tools/ci_build/update_version.py --src "." --version ${VERSION}
+  fi
+
+  prepare_src "$SRCDIR"
+
+  build_wheel "$SRCDIR" "$DSTDIR" "$PKG_NAME_FLAG"
+
+  if [[ $CLEANSRC -ne 0 ]]; then
+    rm -rf "${TMPDIR}"
+  fi
+}
+
+main "$@"
diff --git a/tensorflow_lite_support/tools/pip_package/setup.py b/tensorflow_lite_support/tools/pip_package/setup.py
new file mode 100644
index 0000000..460c505
--- /dev/null
+++ b/tensorflow_lite_support/tools/pip_package/setup.py
@@ -0,0 +1,154 @@
+# lint as: python3
+# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License..
+# ==============================================================================
+"""TFLite Support is a toolkit that helps users to develop ML and deploy TFLite models onto mobile devices.
+
+This PyPI package includes the Python bindings for following features:
+
+ - Metadata schemas: wraps TFLite model schema and metadata schema in Python.
+ - Metadata populator and displayer: can be used to populate the metadata and
+ associated files into the model, as well as converting the populated metadata
+ into the json format.
+ - Android Codegen tool: generates the Java model interface used in Android for
+ a particular model.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import fnmatch
+import os
+import re
+import sys
+
+from setuptools import Command
+from setuptools import find_packages
+from setuptools import setup
+from setuptools.command.install import install as InstallCommandBase
+from setuptools.dist import Distribution
+
+# This version string is semver compatible, but incompatible with pip.
+# For pip, we will remove all '-' characters from this string, and use the
+# result for pip.
+_VERSION = '0.1.0'
+
+SETUP_PACKAGES = [
+    'pybind11 >= 2.6.0',
+]
+
+REQUIRED_PACKAGES = [
+    'absl-py >= 0.7.0',
+    'numpy >= 1.16.0',
+    'flatbuffers >= 1.12',
+] + SETUP_PACKAGES
+
+project_name = 'tflite-support'
+if '--project_name' in sys.argv:
+  project_name_idx = sys.argv.index('--project_name')
+  project_name = sys.argv[project_name_idx + 1]
+  sys.argv.remove('--project_name')
+  sys.argv.pop(project_name_idx)
+
+DOCLINES = __doc__.split('\n')
+
+CONSOLE_SCRIPTS = [
+    'tflite_codegen = tensorflow_lite_support.codegen.python.codegen:main',
+]
+
+
+class BinaryDistribution(Distribution):
+
+  def has_ext_modules(self):
+    return True
+
+
+class InstallCommand(InstallCommandBase):
+  """Override the dir where the headers go."""
+
+  def finalize_options(self):
+    ret = InstallCommandBase.finalize_options(self)
+    self.install_lib = self.install_platlib
+    return ret
+
+
+def find_files(pattern, root):
+  """Return all the files matching pattern below root dir."""
+  for dirpath, _, files in os.walk(root):
+    for filename in fnmatch.filter(files, pattern):
+      yield os.path.join(dirpath, filename)
+
+
+so_lib_paths = [
+    i for i in os.listdir('.')
+    if os.path.isdir(i) and fnmatch.fnmatch(i, '_solib_*')
+]
+
+matches = []
+for path in so_lib_paths:
+  matches.extend(['../' + x for x in find_files('*', path) if '.py' not in x])
+
+EXTENSIONS = ['codegen/_pywrap_codegen.so']
+
+headers = ()
+
+setup(
+    name=project_name,
+    version=_VERSION.replace('-', ''),
+    description=DOCLINES[0],
+    long_description='\n'.join(DOCLINES),
+    long_description_content_type='text/markdown',
+    url='https://www.tensorflow.org/',
+    download_url='https://github.com/tensorflow/tflite-support/tags',
+    author='Google, LLC.',
+    author_email='packages@tensorflow.org',
+    # Contained modules and scripts.
+    packages=find_packages(),
+    entry_points={
+        'console_scripts': CONSOLE_SCRIPTS,
+    },
+    headers=headers,
+    setup_requires=SETUP_PACKAGES,
+    install_requires=REQUIRED_PACKAGES,
+    tests_require=REQUIRED_PACKAGES,
+    # Add in any packaged data.
+    include_package_data=True,
+    package_data={
+        'tflite-support': EXTENSIONS + matches,
+    },
+    zip_safe=False,
+    distclass=BinaryDistribution,
+    cmdclass={
+        'install': InstallCommand,
+    },
+    # PyPI package information.
+    classifiers=sorted([
+        'Development Status :: 3 - Alpha',
+        'Intended Audience :: Developers',
+        'Intended Audience :: Education',
+        'Intended Audience :: Science/Research',
+        'License :: OSI Approved :: Apache Software License',
+        'Programming Language :: Python :: 2.7',
+        'Programming Language :: Python :: 3.6',
+        'Programming Language :: Python :: 3.7',
+        'Programming Language :: Python :: 3.8',
+        'Topic :: Scientific/Engineering',
+        'Topic :: Scientific/Engineering :: Artificial Intelligence',
+        'Topic :: Software Development',
+        'Topic :: Software Development :: Libraries',
+        'Topic :: Software Development :: Libraries :: Python Modules',
+    ]),
+    license='Apache 2.0',
+)
diff --git a/tensorflow_lite_support/tools/pip_package/simple_console_for_windows.py b/tensorflow_lite_support/tools/pip_package/simple_console_for_windows.py
new file mode 100644
index 0000000..106528b
--- /dev/null
+++ b/tensorflow_lite_support/tools/pip_package/simple_console_for_windows.py
@@ -0,0 +1,33 @@
+# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+"""Start a simple interactive console with TensorFlow available."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import code
+import sys
+
+
+def main(_):
+  """Run an interactive console."""
+  code.interact()
+  return 0
+
+
+if __name__ == '__main__':
+  sys.exit(main(sys.argv))
diff --git a/tensorflow_lite_support/tools/pip_package/tflite_support.__init__.py b/tensorflow_lite_support/tools/pip_package/tflite_support.__init__.py
new file mode 100644
index 0000000..c5bf3de
--- /dev/null
+++ b/tensorflow_lite_support/tools/pip_package/tflite_support.__init__.py
@@ -0,0 +1,28 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""An import entry for the TFLite Support project.
+
+In the original project structure, all python targets are accessed by paths like
+tensorflow_lite_support.metadata.metadata.MetadataDisplayer, which is verbose
+and deep. This file provides some shortcuts. It's also compatible with our first
+version Pip package.
+
+In pip build, this file will be renamed as tflite_support/__init__.py.
+"""
+
+import flatbuffers
+from tensorflow_lite_support.metadata import metadata_schema_py_generated
+from tensorflow_lite_support.metadata import schema_py_generated
+from tensorflow_lite_support.metadata.python import metadata
diff --git a/tensorflow_lite_support/tools/zip_files.py b/tensorflow_lite_support/tools/zip_files.py
new file mode 100644
index 0000000..9dc6623
--- /dev/null
+++ b/tensorflow_lite_support/tools/zip_files.py
@@ -0,0 +1,41 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+# Lint as: python3
+"""Creates a zip package of the files passed in."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import zipfile
+
+from absl import app
+from absl import flags
+
+FLAGS = flags.FLAGS
+flags.DEFINE_string("export_zip_path", None, "Path to zip file.")
+flags.DEFINE_string("file_directory", None, "Path to the files to be zipped.")
+
+
+def main(_):
+  with zipfile.ZipFile(FLAGS.export_zip_path, mode="w") as zf:
+    for root, _, files in os.walk(FLAGS.file_directory):
+      for f in files:
+        if f.endswith(".java"):
+          zf.write(os.path.join(root, f))
+
+
+if __name__ == "__main__":
+  app.run(main)
diff --git a/third_party/BUILD b/third_party/BUILD
new file mode 100644
index 0000000..fe756e1
--- /dev/null
+++ b/third_party/BUILD
@@ -0,0 +1 @@
+licenses(["notice"])  # Apache 2.0
diff --git a/third_party/android/BUILD b/third_party/android/BUILD
new file mode 100644
index 0000000..fd69d4b
--- /dev/null
+++ b/third_party/android/BUILD
@@ -0,0 +1 @@
+# Placeholder to make bazel treat it as a package.
diff --git a/third_party/android/android.bzl.tpl b/third_party/android/android.bzl.tpl
new file mode 100644
index 0000000..e6ed499
--- /dev/null
+++ b/third_party/android/android.bzl.tpl
@@ -0,0 +1,9 @@
+"""Set up configurable Android SDK and NDK dependencies."""
+
+def android_workspace():
+  # String for replacement in Bazel template.
+  # These will either be replaced by android_sdk_repository if various ENV
+  # variables are set when `local_config_android` repo_rule is run, or they
+  # will be replaced by noops otherwise.
+  MAYBE_ANDROID_SDK_REPOSITORY
+  MAYBE_ANDROID_NDK_REPOSITORY
diff --git a/third_party/android/android_configure.BUILD.tpl b/third_party/android/android_configure.BUILD.tpl
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/third_party/android/android_configure.BUILD.tpl
diff --git a/third_party/android/android_configure.bzl b/third_party/android/android_configure.bzl
new file mode 100644
index 0000000..2fd2d80
--- /dev/null
+++ b/third_party/android/android_configure.bzl
@@ -0,0 +1,95 @@
+"""Repository rule for Android SDK and NDK autoconfiguration.
+
+`android_configure` depends on the following environment variables:
+
+  * `ANDROID_NDK_HOME`: Location of Android NDK root.
+  * `ANDROID_SDK_HOME`: Location of Android SDK root.
+  * `ANDROID_SDK_API_LEVEL`: Desired Android SDK API version.
+  * `ANDROID_NDK_API_LEVEL`: Desired Android NDK API version.
+  * `ANDROID_BUILD_TOOLS_VERSION`: Desired Android build tools version.
+
+
+Writes Android SDK and NDK rules.
+
+Add the following to your WORKSPACE FILE:
+
+```python
+android_configure(name = "local_config_android")
+```
+
+Args:
+  name: A unique name for this workspace rule.
+"""
+
+_ANDROID_NDK_HOME = "ANDROID_NDK_HOME"
+_ANDROID_SDK_HOME = "ANDROID_SDK_HOME"
+_ANDROID_NDK_API_VERSION = "ANDROID_NDK_API_LEVEL"
+_ANDROID_SDK_API_VERSION = "ANDROID_SDK_API_LEVEL"
+_ANDROID_BUILD_TOOLS_VERSION = "ANDROID_BUILD_TOOLS_VERSION"
+
+_ANDROID_SDK_REPO_TEMPLATE = """
+    native.android_sdk_repository(
+        name="androidsdk",
+        path="%s",
+        api_level=%s,
+        build_tools_version="%s",
+    )
+"""
+
+_ANDROID_NDK_REPO_TEMPLATE = """
+    native.android_ndk_repository(
+        name="androidndk",
+        path="%s",
+        api_level=%s,
+    )
+"""
+
+def _android_autoconf_impl(repository_ctx):
+    """Implementation of the android_autoconf repository rule."""
+    sdk_home = repository_ctx.os.environ.get(_ANDROID_SDK_HOME)
+    sdk_api_level = repository_ctx.os.environ.get(_ANDROID_SDK_API_VERSION)
+    build_tools_version = repository_ctx.os.environ.get(
+        _ANDROID_BUILD_TOOLS_VERSION,
+    )
+    ndk_home = repository_ctx.os.environ.get(_ANDROID_NDK_HOME)
+    ndk_api_level = repository_ctx.os.environ.get(_ANDROID_NDK_API_VERSION)
+
+    sdk_rule = ""
+    if all([sdk_home, sdk_api_level, build_tools_version]):
+        sdk_rule = _ANDROID_SDK_REPO_TEMPLATE % (
+            sdk_home,
+            sdk_api_level,
+            build_tools_version,
+        )
+
+    ndk_rule = ""
+    if all([ndk_home, ndk_api_level]):
+        ndk_rule = _ANDROID_NDK_REPO_TEMPLATE % (ndk_home, ndk_api_level)
+
+    if ndk_rule == "" and sdk_rule == "":
+        sdk_rule = "pass"
+        # TODO(xunkai): Add interactive configure script.
+
+    repository_ctx.template(
+        "BUILD",
+        Label("//third_party/android:android_configure.BUILD.tpl"),
+    )
+    repository_ctx.template(
+        "android.bzl",
+        Label("//third_party/android:android.bzl.tpl"),
+        substitutions = {
+            "MAYBE_ANDROID_SDK_REPOSITORY": sdk_rule,
+            "MAYBE_ANDROID_NDK_REPOSITORY": ndk_rule,
+        },
+    )
+
+android_configure = repository_rule(
+    implementation = _android_autoconf_impl,
+    environ = [
+        _ANDROID_SDK_API_VERSION,
+        _ANDROID_NDK_API_VERSION,
+        _ANDROID_BUILD_TOOLS_VERSION,
+        _ANDROID_NDK_HOME,
+        _ANDROID_SDK_HOME,
+    ],
+)
diff --git a/third_party/com_google_absl.BUILD b/third_party/com_google_absl.BUILD
new file mode 100644
index 0000000..8fca145
--- /dev/null
+++ b/third_party/com_google_absl.BUILD
@@ -0,0 +1,5 @@
+package(default_visibility = ["//visibility:public"])
+
+licenses(["notice"])  # Apache
+
+exports_files(["LICENSE"])
diff --git a/third_party/com_google_absl_f863b622fe13612433fdf43f76547d5edda0c93001.diff b/third_party/com_google_absl_f863b622fe13612433fdf43f76547d5edda0c93001.diff
new file mode 100644
index 0000000..0cd2dff
--- /dev/null
+++ b/third_party/com_google_absl_f863b622fe13612433fdf43f76547d5edda0c93001.diff
@@ -0,0 +1,14 @@
+diff --git a/absl/time/internal/cctz/BUILD.bazel b/absl/time/internal/cctz/BUILD.bazel
+index 9fceffe..e7f9d01 100644
+--- a/absl/time/internal/cctz/BUILD.bazel
++++ b/absl/time/internal/cctz/BUILD.bazel
+@@ -69,8 +69,5 @@ cc_library(
+         "include/cctz/zone_info_source.h",
+     ],
+     linkopts = select({
+-        ":osx": [
+-            "-framework Foundation",
+-        ],
+         ":ios": [
+             "-framework Foundation",
+         ],
\ No newline at end of file
diff --git a/third_party/com_google_protobuf_fixes.diff b/third_party/com_google_protobuf_fixes.diff
new file mode 100644
index 0000000..b9bc17e
--- /dev/null
+++ b/third_party/com_google_protobuf_fixes.diff
@@ -0,0 +1,140 @@
+diff --git a/BUILD b/BUILD
+index 79871d621..51b3a063f 100644
+--- a/BUILD
++++ b/BUILD
+@@ -26,7 +26,7 @@ config_setting(
+ # ZLIB configuration
+ ################################################################################
+
+-ZLIB_DEPS = ["@zlib//:zlib"]
++ZLIB_DEPS = ["@zlib"]
+
+ ################################################################################
+ # Protobuf Runtime Library
+@@ -157,6 +157,7 @@ cc_library(
+     includes = ["src/"],
+     linkopts = LINK_OPTS,
+     visibility = ["//visibility:public"],
++    alwayslink = 1,
+ )
+
+ PROTOBUF_DEPS = select({
+@@ -230,6 +231,7 @@ cc_library(
+     linkopts = LINK_OPTS,
+     visibility = ["//visibility:public"],
+     deps = [":protobuf_lite"] + PROTOBUF_DEPS,
++    alwayslink = 1,
+ )
+
+ # This provides just the header files for use in projects that need to build
+@@ -318,13 +320,13 @@ cc_proto_library(
+
+ [native_cc_proto_library(
+     name = proto + "_cc_proto",
+-    deps = [proto + "_proto"],
+     visibility = ["//visibility:private"],
++    deps = [proto + "_proto"],
+ ) for proto in WELL_KNOWN_PROTO_MAP.keys()]
+
+ cc_proto_blacklist_test(
+     name = "cc_proto_blacklist_test",
+-    deps = [proto + "_cc_proto" for proto in WELL_KNOWN_PROTO_MAP.keys()]
++    deps = [proto + "_cc_proto" for proto in WELL_KNOWN_PROTO_MAP.keys()],
+ )
+
+ ################################################################################
+@@ -900,7 +902,6 @@ py_proto_library(
+     py_extra_srcs = glob(["python/**/__init__.py"]),
+     py_libs = [
+         ":python_srcs",
+-        "@six//:six",
+     ],
+     srcs_version = "PY2AND3",
+     visibility = ["//visibility:public"],
+@@ -1002,7 +1003,9 @@ cc_library(
+ # Note: We use `native_proto_common` here because we depend on an implementation-detail of
+ # `proto_lang_toolchain`, which may not be available on `proto_common`.
+ reject_blacklisted_files = hasattr(native_proto_common, "proto_lang_toolchain_rejects_files_do_not_use_or_we_will_break_you_without_mercy")
++
+ cc_toolchain_blacklisted_protos = [proto + "_proto" for proto in WELL_KNOWN_PROTO_MAP.keys()] if reject_blacklisted_files else [":well_known_protos"]
++
+ proto_lang_toolchain(
+     name = "cc_toolchain",
+     blacklisted_protos = cc_toolchain_blacklisted_protos,
+diff --git a/protobuf.bzl b/protobuf.bzl
+index 829464d44..4ac23594b 100644
+--- a/protobuf.bzl
++++ b/protobuf.bzl
+@@ -87,6 +87,8 @@ def _proto_gen_impl(ctx):
+     for dep in ctx.attr.deps:
+         import_flags += dep.proto.import_flags
+         deps += dep.proto.deps
++    import_flags = depset(import_flags).to_list()
++    deps = depset(deps).to_list()
+
+     if not ctx.attr.gen_cc and not ctx.attr.gen_py and not ctx.executable.plugin:
+         return struct(
+diff --git a/src/google/protobuf/io/gzip_stream.h b/src/google/protobuf/io/gzip_stream.h
+index b1ce1d36c..d5d560ea7 100644
+--- a/src/google/protobuf/io/gzip_stream.h
++++ b/src/google/protobuf/io/gzip_stream.h
+@@ -47,10 +47,12 @@
+ #include <google/protobuf/stubs/common.h>
+ #include <google/protobuf/io/zero_copy_stream.h>
+ #include <google/protobuf/port.h>
+-#include <zlib.h>
+-
+ #include <google/protobuf/port_def.inc>
+
++#if HAVE_ZLIB
++#include <zlib.h>
++#endif  // HAVE_ZLIB
++
+ namespace google {
+ namespace protobuf {
+ namespace io {
+@@ -76,8 +78,10 @@ class PROTOBUF_EXPORT GzipInputStream : public ZeroCopyInputStream {
+   virtual ~GzipInputStream();
+
+   // Return last error message or NULL if no error.
++#if HAVE_ZLIB
+   inline const char* ZlibErrorMessage() const { return zcontext_.msg; }
+   inline int ZlibErrorCode() const { return zerror_; }
++#endif  // HAVE_ZLIB
+
+   // implements ZeroCopyInputStream ----------------------------------
+   bool Next(const void** data, int* size);
+@@ -90,8 +94,10 @@ class PROTOBUF_EXPORT GzipInputStream : public ZeroCopyInputStream {
+
+   ZeroCopyInputStream* sub_stream_;
+
++ #if HAVE_ZLIB
+   z_stream zcontext_;
+   int zerror_;
++ #endif // HAVE_ZLIB
+
+   void* output_buffer_;
+   void* output_position_;
+@@ -142,9 +148,11 @@ class PROTOBUF_EXPORT GzipOutputStream : public ZeroCopyOutputStream {
+
+   virtual ~GzipOutputStream();
+
++#if HAVE_ZLIB
+   // Return last error message or NULL if no error.
+   inline const char* ZlibErrorMessage() const { return zcontext_.msg; }
+   inline int ZlibErrorCode() const { return zerror_; }
++#endif  // HAVE_ZLIB
+
+   // Flushes data written so far to zipped data in the underlying stream.
+   // It is the caller's responsibility to flush the underlying stream if
+@@ -177,8 +185,10 @@ class PROTOBUF_EXPORT GzipOutputStream : public ZeroCopyOutputStream {
+   void* sub_data_;
+   int sub_data_size_;
+
++#if HAVE_ZLIB
+   z_stream zcontext_;
+   int zerror_;
++#endif  //HAVE_ZLIB
+   void* input_buffer_;
+   size_t input_buffer_length_;
+
diff --git a/third_party/darts_clone.BUILD b/third_party/darts_clone.BUILD
new file mode 100644
index 0000000..1d95ec2
--- /dev/null
+++ b/third_party/darts_clone.BUILD
@@ -0,0 +1,15 @@
+# Description:
+#   Darts-clone is a clone of Darts (Double-ARray Trie System).
+
+licenses(["notice"])
+
+exports_files(["LICENSE"])
+
+package(default_visibility = ["//visibility:public"])
+
+cc_library(
+    name = "darts_clone",
+    hdrs = [
+        "include/darts.h",
+    ],
+)
diff --git a/third_party/fft2d/BUILD b/third_party/fft2d/BUILD
new file mode 100644
index 0000000..863a1ce
--- /dev/null
+++ b/third_party/fft2d/BUILD
@@ -0,0 +1,48 @@
+# Headers for 2D Fast Fourier Transform package
+# from http://momonga.t.u-tokyo.ac.jp/~ooura/fft2d.html
+# This is a separate package because the original downloaded archive doesn't
+# contain any header files.
+
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+# Unrestricted use; can only distribute original package.
+# See fft/readme.txt
+licenses(["notice"])
+
+exports_files(["LICENSE"])
+
+cc_library(
+    name = "fft2d_headers",
+    srcs = [
+        "fft.h",
+        "fft2d.h",
+    ],
+)
+
+objc_library(
+    name = "fft2d_headersd_ios",
+    srcs = [
+        "fft.h",
+        "fft2d.h",
+    ],
+)
+
+# Export the source code so that it could be compiled for Andoid native apps.
+filegroup(
+    name = "fft2d_headers_srcs",
+    srcs = [
+        "fft.h",
+        "fft2d.h",
+    ],
+)
+
+filegroup(
+    name = "all_files",
+    srcs = glob(
+        ["**/*"],
+        exclude = ["**/OWNERS"],
+    ),
+    visibility = ["//third_party/tensorflow:__subpackages__"],
+)
diff --git a/third_party/fft2d/LICENSE b/third_party/fft2d/LICENSE
new file mode 100644
index 0000000..2bd8550
--- /dev/null
+++ b/third_party/fft2d/LICENSE
@@ -0,0 +1,3 @@
+Copyright(C) 1997,2001 Takuya OOURA (email: ooura@kurims.kyoto-u.ac.jp).
+You may use, copy, modify this code for any purpose and 
+without fee. You may distribute this ORIGINAL package.
diff --git a/third_party/fft2d/fft.h b/third_party/fft2d/fft.h
new file mode 100644
index 0000000..36d838b
--- /dev/null
+++ b/third_party/fft2d/fft.h
@@ -0,0 +1,36 @@
+/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+// Declarations for 1D FFT routines in third_party/fft2d/fft2d.
+
+#ifndef FFT2D_FFT_H__
+#define FFT2D_FFT_H__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+extern void cdft(int, int, double *, int *, double *);
+extern void rdft(int, int, double *, int *, double *);
+extern void ddct(int, int, double *, int *, double *);
+extern void ddst(int, int, double *, int *, double *);
+extern void dfct(int, double *, double *, int *, double *);
+extern void dfst(int, double *, double *, int *, double *);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // FFT2D_FFT_H__
diff --git a/third_party/fft2d/fft2d.BUILD b/third_party/fft2d/fft2d.BUILD
new file mode 100644
index 0000000..9fa5097
--- /dev/null
+++ b/third_party/fft2d/fft2d.BUILD
@@ -0,0 +1,45 @@
+# 2D Fast Fourier Transform package
+# from http://momonga.t.u-tokyo.ac.jp/~ooura/fft2d.html
+
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+# Unrestricted use; can only distribute original package.
+licenses(["notice"])
+
+exports_files(["readme2d.txt"])
+
+FFT2D_SRCS = [
+    "fftsg.c",
+    "fftsg2d.c",
+]
+
+config_setting(
+    name = "windows",
+    values = {"cpu": "x64_windows"},
+)
+
+# This is the main 2D FFT library.  The 2D FFTs in this library call
+# 1D FFTs.  In addition, fast DCTs are provided for the special case
+# of 8x8 and 16x16.  This code in this library is referred to as
+# "Version II" on http://momonga.t.u-tokyo.ac.jp/~ooura/fft2d.html.
+cc_library(
+    name = "fft2d",
+    srcs = FFT2D_SRCS,
+    linkopts = select({
+        ":windows": [],
+        "//conditions:default": ["-lm"],
+    }),
+)
+
+objc_library(
+    name = "fft2d_ios",
+    srcs = FFT2D_SRCS,
+)
+
+# Export the source code so that it could be compiled for Andoid native apps.
+filegroup(
+    name = "fft2d_srcs",
+    srcs = FFT2D_SRCS,
+)
diff --git a/third_party/fft2d/fft2d.h b/third_party/fft2d/fft2d.h
new file mode 100644
index 0000000..d587b3b
--- /dev/null
+++ b/third_party/fft2d/fft2d.h
@@ -0,0 +1,36 @@
+/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+==============================================================================*/
+
+// Declarations for 2D FFT routines in third_party/fft2d/fft2d.
+
+#ifndef FFT2D_FFT_H__
+#define FFT2D_FFT_H__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+extern void cdft2d(int, int, int, double **, double *, int *, double *);
+extern void rdft2d(int, int, int, double **, double *, int *, double *);
+extern void ddct2d(int, int, int, double **, double *, int *, double *);
+extern void ddst2d(int, int, int, double **, double *, int *, double *);
+extern void ddct8x8s(int isgn, double **a);
+extern void ddct16x16s(int isgn, double **a);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // FFT2D_FFT_H__
diff --git a/third_party/flatbuffers/BUILD b/third_party/flatbuffers/BUILD
new file mode 100644
index 0000000..82bab3f
--- /dev/null
+++ b/third_party/flatbuffers/BUILD
@@ -0,0 +1 @@
+# This empty BUILD file is required to make Bazel treat this directory as a package.
diff --git a/third_party/flatbuffers/BUILD.bazel b/third_party/flatbuffers/BUILD.bazel
new file mode 100644
index 0000000..1ee46f0
--- /dev/null
+++ b/third_party/flatbuffers/BUILD.bazel
@@ -0,0 +1,140 @@
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+
+package(default_visibility = ["//visibility:public"])
+
+licenses(["notice"])  # Apache 2.0
+
+exports_files(["LICENSE.txt"])
+
+licenses(["notice"])
+
+config_setting(
+    name = "freebsd",
+    values = {"cpu": "freebsd"},
+)
+
+config_setting(
+    name = "windows",
+    values = {"cpu": "x64_windows"},
+)
+
+load("@rules_cc//cc:defs.bzl", "cc_binary", "cc_library")
+
+# Public flatc library to compile flatbuffer files at runtime.
+cc_library(
+    name = "flatbuffers",
+    hdrs = ["//:public_headers"],
+    linkstatic = 1,
+    strip_include_prefix = "/include",
+    visibility = ["//visibility:public"],
+    deps = ["//src:flatbuffers"],
+)
+
+# Public C++ headers for the Flatbuffers library.
+filegroup(
+    name = "public_headers",
+    srcs = [
+        "include/flatbuffers/base.h",
+        "include/flatbuffers/code_generators.h",
+        "include/flatbuffers/flatbuffers.h",
+        "include/flatbuffers/flexbuffers.h",
+        "include/flatbuffers/hash.h",
+        "include/flatbuffers/idl.h",
+        "include/flatbuffers/minireflect.h",
+        "include/flatbuffers/reflection.h",
+        "include/flatbuffers/reflection_generated.h",
+        "include/flatbuffers/registry.h",
+        "include/flatbuffers/stl_emulation.h",
+        "include/flatbuffers/util.h",
+    ],
+    visibility = ["//:__subpackages__"],
+)
+
+# Public flatc compiler library.
+cc_library(
+    name = "flatc_library",
+    linkstatic = 1,
+    visibility = ["//visibility:public"],
+    deps = [
+        "@flatbuffers//src:flatc_library",
+    ],
+)
+
+# Public flatc compiler.
+cc_binary(
+    name = "flatc",
+    linkopts = select({
+        ":freebsd": [
+            "-lm",
+        ],
+        ":windows": [],
+        "//conditions:default": [
+            "-lm",
+            "-ldl",
+        ],
+    }),
+    visibility = ["//visibility:public"],
+    deps = [
+        "@flatbuffers//src:flatc",
+    ],
+)
+
+filegroup(
+    name = "flatc_headers",
+    srcs = [
+        "include/flatbuffers/flatc.h",
+    ],
+    visibility = ["//:__subpackages__"],
+)
+
+# Library used by flatbuffer_cc_library rules.
+cc_library(
+    name = "runtime_cc",
+    hdrs = [
+        "include/flatbuffers/base.h",
+        "include/flatbuffers/flatbuffers.h",
+        "include/flatbuffers/flexbuffers.h",
+        "include/flatbuffers/stl_emulation.h",
+        "include/flatbuffers/util.h",
+    ],
+    linkstatic = 1,
+    strip_include_prefix = "/include",
+    visibility = ["//visibility:public"],
+)
+
+filegroup(
+    name = "runtime_py_srcs",
+    srcs = [
+        "python/flatbuffers/__init__.py",
+        "python/flatbuffers/builder.py",
+        "python/flatbuffers/compat.py",
+        "python/flatbuffers/encode.py",
+        "python/flatbuffers/number_types.py",
+        "python/flatbuffers/packer.py",
+        "python/flatbuffers/table.py",
+        "python/flatbuffers/util.py",
+    ],
+)
+
+py_library(
+    name = "runtime_py",
+    srcs = [":runtime_py_srcs"],
+    visibility = ["//visibility:public"],
+)
+
+filegroup(
+    name = "runtime_java_srcs",
+    srcs = glob(["java/com/google/flatbuffers/**/*.java"]),
+)
+
+java_library(
+    name = "runtime_java",
+    srcs = [":runtime_java_srcs"],
+    visibility = ["//visibility:public"],
+)
+
+android_library(
+    name = "runtime_android",
+    srcs = [":runtime_java_srcs"],
+    visibility = ["//visibility:public"],
+)
diff --git a/third_party/flatbuffers/build_defs.bzl b/third_party/flatbuffers/build_defs.bzl
new file mode 100644
index 0000000..986b8d1
--- /dev/null
+++ b/third_party/flatbuffers/build_defs.bzl
@@ -0,0 +1,617 @@
+"""BUILD rules for generating flatbuffer files."""
+
+load("@build_bazel_rules_android//android:rules.bzl", "android_library")
+
+flatc_path = "@flatbuffers//:flatc"
+zip_files = "//tensorflow_lite_support/tools:zip_files"
+
+DEFAULT_INCLUDE_PATHS = [
+    "./",
+    "$(GENDIR)",
+    "$(BINDIR)",
+]
+
+DEFAULT_FLATC_ARGS = [
+    "--no-union-value-namespacing",
+    "--gen-object-api",
+]
+
+def flatbuffer_library_public(
+        name,
+        srcs,
+        outs,
+        language_flag,
+        out_prefix = "",
+        includes = [],
+        include_paths = [],
+        compatible_with = [],
+        flatc_args = DEFAULT_FLATC_ARGS,
+        reflection_name = "",
+        reflection_visibility = None,
+        output_to_bindir = False):
+    """Generates code files for reading/writing the given flatbuffers in the requested language using the public compiler.
+
+    Outs:
+      filegroup(name): all generated source files.
+      Fileset([reflection_name]): (Optional) all generated reflection binaries.
+
+    Args:
+      name: Rule name.
+      srcs: Source .fbs files. Sent in order to the compiler.
+      outs: Output files from flatc.
+      language_flag: Target language flag. One of [-c, -j, -js].
+      out_prefix: Prepend this path to the front of all generated files except on
+          single source targets. Usually is a directory name.
+      includes: Optional, list of filegroups of schemas that the srcs depend on.
+      include_paths: Optional, list of paths the includes files can be found in.
+      compatible_with: Optional, passed to genrule for environments this rule
+          can be built for.
+      flatc_args: Optional, list of additional arguments to pass to flatc.
+      reflection_name: Optional, if set this will generate the flatbuffer
+        reflection binaries for the schemas.
+      reflection_visibility: The visibility of the generated reflection Fileset.
+      output_to_bindir: Passed to genrule for output to bin directory.
+    """
+    include_paths_cmd = ["-I %s" % (s) for s in include_paths]
+
+    # '$(@D)' when given a single source target will give the appropriate
+    # directory. Appending 'out_prefix' is only necessary when given a build
+    # target with multiple sources.
+    output_directory = (
+        ("-o $(@D)/%s" % (out_prefix)) if len(srcs) > 1 else ("-o $(@D)")
+    )
+    genrule_cmd = " ".join([
+        "for f in $(SRCS); do",
+        "$(location %s)" % (flatc_path),
+        " ".join(flatc_args),
+        " ".join(include_paths_cmd),
+        language_flag,
+        output_directory,
+        "$$f;",
+        "done",
+    ])
+    native.genrule(
+        name = name,
+        srcs = srcs,
+        outs = outs,
+        output_to_bindir = output_to_bindir,
+        compatible_with = compatible_with,
+        tools = includes + [flatc_path],
+        cmd = genrule_cmd,
+        message = "Generating flatbuffer files for %s:" % (name),
+    )
+    if reflection_name:
+        reflection_genrule_cmd = " ".join([
+            "for f in $(SRCS); do",
+            "$(location %s)" % (flatc_path),
+            "-b --schema",
+            " ".join(flatc_args),
+            " ".join(include_paths_cmd),
+            language_flag,
+            output_directory,
+            "$$f;",
+            "done",
+        ])
+        reflection_outs = [
+            (out_prefix + "%s.bfbs") % (s.replace(".fbs", "").split("/")[-1])
+            for s in srcs
+        ]
+        native.genrule(
+            name = "%s_srcs" % reflection_name,
+            srcs = srcs,
+            outs = reflection_outs,
+            output_to_bindir = output_to_bindir,
+            compatible_with = compatible_with,
+            tools = includes + [flatc_path],
+            cmd = reflection_genrule_cmd,
+            message = "Generating flatbuffer reflection binary for %s:" % (name),
+        )
+        # TODO(b/114456773): Make bazel rules proper and supported by flatbuffer
+        # Have to comment this since FilesetEntry is not supported in bazel
+        # skylark.
+        # native.Fileset(
+        #     name = reflection_name,
+        #     out = "%s_out" % reflection_name,
+        #     entries = [
+        #         native.FilesetEntry(files = reflection_outs),
+        #     ],
+        #     visibility = reflection_visibility,
+        #     compatible_with = compatible_with,
+        # )
+
+def flatbuffer_cc_library(
+        name,
+        srcs,
+        srcs_filegroup_name = "",
+        out_prefix = "",
+        includes = [],
+        include_paths = [],
+        compatible_with = [],
+        flatc_args = DEFAULT_FLATC_ARGS,
+        visibility = None,
+        srcs_filegroup_visibility = None,
+        gen_reflections = False):
+    '''A cc_library with the generated reader/writers for the given flatbuffer definitions.
+
+    Outs:
+      filegroup([name]_srcs): all generated .h files.
+      filegroup(srcs_filegroup_name if specified, or [name]_includes if not):
+          Other flatbuffer_cc_library's can pass this in for their `includes`
+          parameter, if they depend on the schemas in this library.
+      Fileset([name]_reflection): (Optional) all generated reflection binaries.
+      cc_library([name]): library with sources and flatbuffers deps.
+
+    Remarks:
+      ** Because the genrule used to call flatc does not have any trivial way of
+        computing the output list of files transitively generated by includes and
+        --gen-includes (the default) being defined for flatc, the --gen-includes
+        flag will not work as expected. The way around this is to add a dependency
+        to the flatbuffer_cc_library defined alongside the flatc included Fileset.
+        For example you might define:
+
+        flatbuffer_cc_library(
+            name = "my_fbs",
+            srcs = [ "schemas/foo.fbs" ],
+            includes = [ "//third_party/bazz:bazz_fbs_includes" ],
+        )
+
+        In which foo.fbs includes a few files from the Fileset defined at
+        //third_party/bazz:bazz_fbs_includes. When compiling the library that
+        includes foo_generated.h, and therefore has my_fbs as a dependency, it
+        will fail to find any of the bazz *_generated.h files unless you also
+        add bazz's flatbuffer_cc_library to your own dependency list, e.g.:
+
+        cc_library(
+            name = "my_lib",
+            deps = [
+                ":my_fbs",
+                "//third_party/bazz:bazz_fbs"
+            ],
+        )
+
+        Happy dependent Flatbuffering!
+
+    Args:
+      name: Rule name.
+      srcs: Source .fbs files. Sent in order to the compiler.
+      srcs_filegroup_name: Name of the output filegroup that holds srcs. Pass this
+          filegroup into the `includes` parameter of any other
+          flatbuffer_cc_library that depends on this one's schemas.
+      out_prefix: Prepend this path to the front of all generated files. Usually
+          is a directory name.
+      includes: Optional, list of filegroups of schemas that the srcs depend on.
+          ** SEE REMARKS BELOW **
+      include_paths: Optional, list of paths the includes files can be found in.
+      compatible_with: Optional, passed to genrule for environments this rule
+          can be built for
+      flatc_args: Optional list of additional arguments to pass to flatc
+          (e.g. --gen-mutable).
+      visibility: The visibility of the generated cc_library. By default, use the
+          default visibility of the project.
+      srcs_filegroup_visibility: The visibility of the generated srcs filegroup.
+          By default, use the value of the visibility parameter above.
+      gen_reflections: Optional, if true this will generate the flatbuffer
+        reflection binaries for the schemas.
+    '''
+    output_headers = [
+        (out_prefix + "%s_generated.h") % (s.replace(".fbs", "").split("/")[-1])
+        for s in srcs
+    ]
+    reflection_name = "%s_reflection" % name if gen_reflections else ""
+
+    flatbuffer_library_public(
+        name = "%s_srcs" % (name),
+        srcs = srcs,
+        outs = output_headers,
+        language_flag = "-c",
+        out_prefix = out_prefix,
+        includes = includes,
+        include_paths = include_paths,
+        compatible_with = compatible_with,
+        flatc_args = flatc_args,
+        reflection_name = reflection_name,
+        reflection_visibility = visibility,
+    )
+    native.cc_library(
+        name = name,
+        hdrs = output_headers,
+        srcs = output_headers,
+        features = [
+            "-parse_headers",
+        ],
+        deps = [
+            "@flatbuffers//:runtime_cc",
+        ],
+        includes = ["."],
+        linkstatic = 1,
+        visibility = visibility,
+        compatible_with = compatible_with,
+    )
+
+    # A filegroup for the `srcs`. That is, all the schema files for this
+    # Flatbuffer set.
+    native.filegroup(
+        name = srcs_filegroup_name if srcs_filegroup_name else "%s_includes" % (name),
+        srcs = srcs,
+        visibility = srcs_filegroup_visibility if srcs_filegroup_visibility != None else visibility,
+        compatible_with = compatible_with,
+    )
+
+# Custom provider to track dependencies transitively.
+FlatbufferInfo = provider(
+    fields = {
+        "transitive_srcs": "flatbuffer schema definitions.",
+    },
+)
+
+def _flatbuffer_schemas_aspect_impl(target, ctx):
+    _ignore = [target]
+    transitive_srcs = depset()
+    if hasattr(ctx.rule.attr, "deps"):
+        for dep in ctx.rule.attr.deps:
+            if FlatbufferInfo in dep:
+                transitive_srcs = depset(dep[FlatbufferInfo].transitive_srcs, transitive = [transitive_srcs])
+    if hasattr(ctx.rule.attr, "srcs"):
+        for src in ctx.rule.attr.srcs:
+            if FlatbufferInfo in src:
+                transitive_srcs = depset(src[FlatbufferInfo].transitive_srcs, transitive = [transitive_srcs])
+            for f in src.files:
+                if f.extension == "fbs":
+                    transitive_srcs = depset([f], transitive = [transitive_srcs])
+    return [FlatbufferInfo(transitive_srcs = transitive_srcs)]
+
+# An aspect that runs over all dependencies and transitively collects
+# flatbuffer schema files.
+_flatbuffer_schemas_aspect = aspect(
+    attr_aspects = [
+        "deps",
+        "srcs",
+    ],
+    implementation = _flatbuffer_schemas_aspect_impl,
+)
+
+# Rule to invoke the flatbuffer compiler.
+def _gen_flatbuffer_srcs_impl(ctx):
+    outputs = ctx.attr.outputs
+    include_paths = ctx.attr.include_paths
+    if ctx.attr.no_includes:
+        no_includes_statement = ["--no-includes"]
+    else:
+        no_includes_statement = []
+
+    # Need to generate all files in a directory.
+    if not outputs:
+        outputs = [ctx.actions.declare_directory("{}_all".format(ctx.attr.name))]
+        output_directory = outputs[0].path
+    else:
+        outputs = [ctx.actions.declare_file(output) for output in outputs]
+        output_directory = outputs[0].dirname
+
+    deps = depset(ctx.files.srcs + ctx.files.deps, transitive = [
+        dep[FlatbufferInfo].transitive_srcs
+        for dep in ctx.attr.deps
+        if FlatbufferInfo in dep
+    ])
+
+    include_paths_cmd_line = []
+    for s in include_paths:
+        include_paths_cmd_line.extend(["-I", s])
+
+    for src in ctx.files.srcs:
+        ctx.actions.run(
+            inputs = deps,
+            outputs = outputs,
+            executable = ctx.executable._flatc,
+            arguments = [
+                            ctx.attr.language_flag,
+                            "-o",
+                            output_directory,
+                            # Allow for absolute imports and referencing of generated files.
+                            "-I",
+                            "./",
+                            "-I",
+                            ctx.genfiles_dir.path,
+                            "-I",
+                            ctx.bin_dir.path,
+                        ] + no_includes_statement +
+                        include_paths_cmd_line + [
+                "--no-union-value-namespacing",
+                "--gen-object-api",
+                src.path,
+            ],
+            progress_message = "Generating flatbuffer files for {}:".format(src),
+        )
+    return [
+        DefaultInfo(files = depset(outputs)),
+    ]
+
+_gen_flatbuffer_srcs = rule(
+    _gen_flatbuffer_srcs_impl,
+    attrs = {
+        "srcs": attr.label_list(
+            allow_files = [".fbs"],
+            mandatory = True,
+        ),
+        "outputs": attr.string_list(
+            default = [],
+            mandatory = False,
+        ),
+        "deps": attr.label_list(
+            default = [],
+            mandatory = False,
+            aspects = [_flatbuffer_schemas_aspect],
+        ),
+        "include_paths": attr.string_list(
+            default = [],
+            mandatory = False,
+        ),
+        "language_flag": attr.string(
+            mandatory = True,
+        ),
+        "no_includes": attr.bool(
+            default = False,
+            mandatory = False,
+        ),
+        "_flatc": attr.label(
+            default = Label("@flatbuffers//:flatc"),
+            executable = True,
+            cfg = "host",
+        ),
+    },
+    output_to_genfiles = True,
+)
+
+def _concat_flatbuffer_py_srcs_impl(ctx):
+    # Merge all generated python files. The files are concatenated and the
+    # import statements are removed. Finally we import the flatbuffer runtime
+    # library.
+    command = "echo 'import flatbuffers\n' > %s; "
+    command += "for f in $(find %s -name '*.py'); do cat $f | sed '/import flatbuffers/d' >> %s; done "
+    ctx.actions.run_shell(
+        inputs = ctx.attr.deps[0].files,
+        outputs = [ctx.outputs.out],
+        command = command % (
+            ctx.outputs.out.path,
+            ctx.attr.deps[0].files.to_list()[0].path,
+            ctx.outputs.out.path,
+        ),
+    )
+
+_concat_flatbuffer_py_srcs = rule(
+    _concat_flatbuffer_py_srcs_impl,
+    attrs = {
+        "deps": attr.label_list(mandatory = True),
+    },
+    output_to_genfiles = True,
+    outputs = {"out": "%{name}.py"},
+)
+
+def flatbuffer_py_library(
+        name,
+        srcs,
+        deps = [],
+        include_paths = []):
+    """A py_library with the generated reader/writers for the given schema.
+
+    This rule assumes that the schema files define non-conflicting names, so that
+    they can be merged in a single file. This is e.g. the case if only a single
+    namespace is used.
+    The rule call the flatbuffer compiler for all schema files and merges the
+    generated python files into a single file that is wrapped in a py_library.
+
+    Args:
+      name: Rule name. (required)
+      srcs: List of source .fbs files. (required)
+      deps: List of dependencies.
+      include_paths: Optional, list of paths the includes files can be found in.
+    """
+    all_srcs = "{}_srcs".format(name)
+    _gen_flatbuffer_srcs(
+        name = all_srcs,
+        srcs = srcs,
+        language_flag = "--python",
+        deps = deps,
+        include_paths = include_paths,
+    )
+    all_srcs_no_include = "{}_srcs_no_include".format(name)
+    _gen_flatbuffer_srcs(
+        name = all_srcs_no_include,
+        srcs = srcs,
+        language_flag = "--python",
+        deps = deps,
+        no_includes = True,
+        include_paths = include_paths,
+    )
+    concat_py_srcs = "{}_generated".format(name)
+    _concat_flatbuffer_py_srcs(
+        name = concat_py_srcs,
+        deps = [
+            ":{}".format(all_srcs_no_include),
+        ],
+    )
+    native.py_library(
+        name = name,
+        srcs = [
+            ":{}".format(concat_py_srcs),
+        ],
+        srcs_version = "PY2AND3",
+        deps = deps,
+    )
+
+def flatbuffer_java_library(
+        name,
+        srcs,
+        custom_package = "",
+        package_prefix = "",
+        include_paths = DEFAULT_INCLUDE_PATHS,
+        flatc_args = DEFAULT_FLATC_ARGS,
+        visibility = None):
+    """A java library with the generated reader/writers for the given flatbuffer definitions.
+
+    Args:
+      name: Rule name. (required)
+      srcs: List of source .fbs files including all includes. (required)
+      custom_package: Package name of generated Java files. If not specified
+          namespace in the schema files will be used. (optional)
+      package_prefix: like custom_package, but prefixes to the existing
+          namespace. (optional)
+      include_paths: List of paths that includes files can be found in. (optional)
+      flatc_args: List of additional arguments to pass to flatc. (optional)
+      visibility: Visibility setting for the java_library rule. (optional)
+    """
+    out_srcjar = "java_%s_all.srcjar" % name
+    flatbuffer_java_srcjar(
+        name = "%s_srcjar" % name,
+        srcs = srcs,
+        out = out_srcjar,
+        custom_package = custom_package,
+        flatc_args = flatc_args,
+        include_paths = include_paths,
+        package_prefix = package_prefix,
+    )
+
+    native.filegroup(
+        name = "%s.srcjar" % name,
+        srcs = [out_srcjar],
+    )
+
+    native.java_library(
+        name = name,
+        srcs = [out_srcjar],
+        javacopts = ["-source 7 -target 7"],
+        deps = [
+            "@flatbuffers//:runtime_java",
+        ],
+        visibility = visibility,
+    )
+
+def flatbuffer_java_srcjar(
+        name,
+        srcs,
+        out,
+        custom_package = "",
+        package_prefix = "",
+        include_paths = DEFAULT_INCLUDE_PATHS,
+        flatc_args = DEFAULT_FLATC_ARGS):
+    """Generate flatbuffer Java source files.
+
+    Args:
+      name: Rule name. (required)
+      srcs: List of source .fbs files including all includes. (required)
+      out: Output file name. (required)
+      custom_package: Package name of generated Java files. If not specified
+          namespace in the schema files will be used. (optional)
+      package_prefix: like custom_package, but prefixes to the existing
+          namespace. (optional)
+      include_paths: List of paths that includes files can be found in. (optional)
+      flatc_args: List of additional arguments to pass to flatc. (optional)
+    """
+    command_fmt = """set -e
+      tmpdir=$(@D)
+      schemas=$$tmpdir/schemas
+      java_root=$$tmpdir/java
+      rm -rf $$schemas
+      rm -rf $$java_root
+      mkdir -p $$schemas
+      mkdir -p $$java_root
+
+      for src in $(SRCS); do
+        dest=$$schemas/$$src
+        rm -rf $$(dirname $$dest)
+        mkdir -p $$(dirname $$dest)
+        if [ -z "{custom_package}" ] && [ -z "{package_prefix}" ]; then
+          cp -f $$src $$dest
+        else
+          if [ -z "{package_prefix}" ]; then
+            sed -e "s/namespace\\s.*/namespace {custom_package};/" $$src > $$dest
+          else
+            sed -e "s/namespace \\([^;]\\+\\);/namespace {package_prefix}.\\1;/" $$src > $$dest
+          fi
+        fi
+      done
+
+      flatc_arg_I="-I $$tmpdir/schemas"
+      for include_path in {include_paths}; do
+        flatc_arg_I="$$flatc_arg_I -I $$schemas/$$include_path"
+      done
+
+      flatc_additional_args=
+      for arg in {flatc_args}; do
+        flatc_additional_args="$$flatc_additional_args $$arg"
+      done
+
+      for src in $(SRCS); do
+        $(location {flatc_path}) $$flatc_arg_I --java $$flatc_additional_args -o $$java_root  $$schemas/$$src
+      done
+
+      $(location {zip_files}) -export_zip_path=$@ -file_directory=$$java_root
+      """
+    genrule_cmd = command_fmt.format(
+        package_name = native.package_name(),
+        custom_package = custom_package,
+        package_prefix = package_prefix,
+        flatc_path = flatc_path,
+        zip_files = zip_files,
+        include_paths = " ".join(include_paths),
+        flatc_args = " ".join(flatc_args),
+    )
+
+    native.genrule(
+        name = name,
+        srcs = srcs,
+        outs = [out],
+        tools = [flatc_path, zip_files],
+        cmd = genrule_cmd,
+    )
+
+def flatbuffer_android_library(
+        name,
+        srcs,
+        custom_package = "",
+        package_prefix = "",
+        include_paths = DEFAULT_INCLUDE_PATHS,
+        flatc_args = DEFAULT_FLATC_ARGS,
+        visibility = None):
+    """An android_library with the generated reader/writers for the given flatbuffer definitions.
+
+    Args:
+      name: Rule name. (required)
+      srcs: List of source .fbs files including all includes. (required)
+      custom_package: Package name of generated Java files. If not specified
+          namespace in the schema files will be used. (optional)
+      package_prefix: like custom_package, but prefixes to the existing
+          namespace. (optional)
+      include_paths: List of paths that includes files can be found in. (optional)
+      flatc_args: List of additional arguments to pass to flatc. (optional)
+      visibility: Visibility setting for the android_library rule. (optional)
+    """
+    out_srcjar = "android_%s_all.srcjar" % name
+    flatbuffer_java_srcjar(
+        name = "%s_srcjar" % name,
+        srcs = srcs,
+        out = out_srcjar,
+        custom_package = custom_package,
+        flatc_args = flatc_args,
+        include_paths = include_paths,
+        package_prefix = package_prefix,
+    )
+
+    native.filegroup(
+        name = "%s.srcjar" % name,
+        srcs = [out_srcjar],
+    )
+
+    # To support org.checkerframework.dataflow.qual.Pure.
+    checkerframework_annotations = [
+        "@org_checkerframework_qual",
+    ] if "--java-checkerframework" in flatc_args else []
+
+    android_library(
+        name = name,
+        srcs = [out_srcjar],
+        javacopts = ["-source 7 -target 7"],
+        visibility = visibility,
+        deps = [
+            "@flatbuffers//:runtime_android",
+        ] + checkerframework_annotations,
+    )
diff --git a/third_party/flatbuffers/workspace.bzl b/third_party/flatbuffers/workspace.bzl
new file mode 100644
index 0000000..dea463f
--- /dev/null
+++ b/third_party/flatbuffers/workspace.bzl
@@ -0,0 +1,19 @@
+"""Loads the Flatbuffers library, used by TF Lite."""
+
+load("//third_party:repo.bzl", "third_party_http_archive")
+
+def repo():
+    third_party_http_archive(
+        name = "flatbuffers",
+        strip_prefix = "flatbuffers-1.12.0",
+        sha256 = "62f2223fb9181d1d6338451375628975775f7522185266cd5296571ac152bc45",
+        urls = [
+            "https://storage.googleapis.com/mirror.tensorflow.org/github.com/google/flatbuffers/archive/v1.12.0.tar.gz",
+            "https://github.com/google/flatbuffers/archive/v1.12.0.tar.gz",
+        ],
+        build_file = "//third_party/flatbuffers:BUILD.bazel",
+        delete = ["build_defs.bzl"],
+        link_files = {
+            "//third_party/flatbuffers:build_defs.bzl": "build_defs.bzl",
+        },
+    )
diff --git a/third_party/gflags/BUILD b/third_party/gflags/BUILD
new file mode 100644
index 0000000..82bab3f
--- /dev/null
+++ b/third_party/gflags/BUILD
@@ -0,0 +1 @@
+# This empty BUILD file is required to make Bazel treat this directory as a package.
diff --git a/third_party/gflags/fix_android_pthread_link.patch b/third_party/gflags/fix_android_pthread_link.patch
new file mode 100644
index 0000000..9a0b351
--- /dev/null
+++ b/third_party/gflags/fix_android_pthread_link.patch
@@ -0,0 +1,32 @@
+diff --git a/BUILD b/BUILD
+index 0a5c9eb..d836578 100644
+--- a/BUILD
++++ b/BUILD
+@@ -6,6 +6,11 @@ licenses(["notice"])
+
+ exports_files(["src/gflags_complections.sh", "COPYING.txt"])
+
++config_setting(
++    name = "android",
++    values = {"crosstool_top": "//external:android/crosstool"},
++)
++
+ load(":bazel/gflags.bzl", "gflags_sources", "gflags_library")
+ (hdrs, srcs) = gflags_sources(namespace=["gflags", "google"])
+ gflags_library(hdrs=hdrs, srcs=srcs, threads=0)
+diff --git a/bazel/gflags.bzl b/bazel/gflags.bzl
+index cd0edad..5c1d8b5 100644
+--- a/bazel/gflags.bzl
++++ b/bazel/gflags.bzl
+@@ -77,7 +77,10 @@ def gflags_library(hdrs=[], srcs=[], threads=1):
+     ]
+     linkopts = []
+     if threads:
+-        linkopts.append("-lpthread")
++        linkopts += select({
++            "//:android": [],
++            "//conditions:default": ["-lpthread"],
++        })
+     else:
+         name += "_nothreads"
+         copts.append("-DNO_THREADS")
\ No newline at end of file
diff --git a/third_party/gflags/workspace.bzl b/third_party/gflags/workspace.bzl
new file mode 100644
index 0000000..194a9d3
--- /dev/null
+++ b/third_party/gflags/workspace.bzl
@@ -0,0 +1,16 @@
+"""Loads the GFlags repo and patch it with android linkopt fix."""
+
+load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")
+
+def repo():
+    http_archive(
+        name = "com_github_gflags_gflags",
+        sha256 = "ae27cdbcd6a2f935baa78e4f21f675649271634c092b1be01469440495609d0e",
+        strip_prefix = "gflags-2.2.1",
+        urls = [
+            "http://mirror.tensorflow.org/github.com/gflags/gflags/archive/v2.2.1.tar.gz",
+            "https://github.com/gflags/gflags/archive/v2.2.1.tar.gz",
+        ],
+        patches = ["@//third_party/gflags:fix_android_pthread_link.patch"],
+        patch_args = ["-p1"],
+    )
diff --git a/third_party/google_toolbox_for_mac.BUILD b/third_party/google_toolbox_for_mac.BUILD
new file mode 100644
index 0000000..8d7fecf
--- /dev/null
+++ b/third_party/google_toolbox_for_mac.BUILD
@@ -0,0 +1,22 @@
+# Description:
+# A collection of source from different Google projects that may be of use to
+# developers working other Mac projects.
+package(
+    default_visibility = ["//visibility:private"],
+)
+
+licenses(["notice"])  # Apache 2.0
+
+exports_files(["LICENSE"])
+
+exports_files(
+    ["UnitTest-Info.plist"],
+    visibility = ["//visibility:public"],
+)
+
+objc_library(
+    name = "GTM_Defines",
+    hdrs = ["GTMDefines.h"],
+    includes = ["."],
+    visibility = ["//visibility:public"],
+)
diff --git a/third_party/icu.BUILD b/third_party/icu.BUILD
new file mode 100644
index 0000000..7749dda
--- /dev/null
+++ b/third_party/icu.BUILD
@@ -0,0 +1,97 @@
+"""Builds ICU library."""
+
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+licenses(["notice"])  # Apache 2.0
+
+exports_files([
+    "icu4c/LICENSE",
+    "icu4j/main/shared/licenses/LICENSE",
+])
+
+cc_library(
+    name = "headers",
+    hdrs = glob(["icu4c/source/common/unicode/*.h"]),
+    includes = [
+        "icu4c/source/common",
+    ],
+    deps = [
+    ],
+)
+
+cc_library(
+    name = "common",
+    hdrs = glob(["icu4c/source/common/unicode/*.h"]),
+    includes = [
+        "icu4c/source/common",
+    ],
+    deps = [
+        ":icuuc",
+    ],
+)
+
+alias(
+    name = "nfkc",
+    actual = ":common",
+)
+
+alias(
+    name = "nfkc_cf",
+    actual = ":common",
+)
+
+cc_library(
+    name = "icuuc",
+    srcs = glob(
+        [
+            "icu4c/source/common/*.c",
+            "icu4c/source/common/*.cpp",
+            "icu4c/source/stubdata/*.cpp",
+        ],
+    ),
+    hdrs = glob([
+        "icu4c/source/common/*.h",
+    ]),
+    copts = [
+        "-DU_COMMON_IMPLEMENTATION",
+    ] + select({
+        ":android": [
+            "-fdata-sections",
+            "-DU_HAVE_NL_LANGINFO_CODESET=0",
+            "-Wno-deprecated-declarations",
+        ],
+        ":apple": [
+            "-Wno-shorten-64-to-32",
+            "-Wno-unused-variable",
+        ],
+        ":windows": [
+            "/utf-8",
+            "/DLOCALE_ALLOW_NEUTRAL_NAMES=0",
+        ],
+        "//conditions:default": [],
+    }),
+    tags = ["requires-rtti"],
+    visibility = [
+        "//visibility:private",
+    ],
+    deps = [
+        ":headers",
+    ],
+)
+
+config_setting(
+    name = "android",
+    values = {"crosstool_top": "//external:android/crosstool"},
+)
+
+config_setting(
+    name = "apple",
+    values = {"cpu": "darwin"},
+)
+
+config_setting(
+    name = "windows",
+    values = {"cpu": "x64_windows"},
+)
diff --git a/third_party/libyuv.BUILD b/third_party/libyuv.BUILD
new file mode 100644
index 0000000..4b39a8c
--- /dev/null
+++ b/third_party/libyuv.BUILD
@@ -0,0 +1,25 @@
+# Description:
+#   The libyuv package provides implementation yuv image conversion, rotation
+#   and scaling.
+
+licenses(["notice"])  # BSD license
+
+exports_files(["LICENSE"])
+
+cc_library(
+    name = "libyuv",
+    srcs = glob(
+        [
+            "source/*.cc",
+            "include/libyuv/*.h",
+        ],
+    ),
+    hdrs = [
+        "include/libyuv.h",
+        "include/libyuv/compare.h",
+        "include/libyuv/convert.h",
+        "include/libyuv/video_common.h",
+    ],
+    includes = ["include"],
+    visibility = ["//visibility:public"],
+)
diff --git a/third_party/libzip.BUILD b/third_party/libzip.BUILD
new file mode 100644
index 0000000..b69ccf4
--- /dev/null
+++ b/third_party/libzip.BUILD
@@ -0,0 +1,189 @@
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+load("@org_tensorflow_lite_support//tensorflow_lite_support/tools:build_rules/expand_template.bzl", "cmake_substitutions", "expand_template")
+
+_CMAKE_VARIABLES = {
+    "INT16_T_LIBZIP": 2,
+    "INT32_T_LIBZIP": 4,
+    "INT64_T_LIBZIP": 8,
+    "INT8_T_LIBZIP": 1,
+    "INT_LIBZIP": 4,
+    "LIBZIP_TYPES_INCLUDE": "#include <stdint.h>",
+    "LONG_LIBZIP": 8,
+    "LONG_LONG_LIBZIP": 8,
+    "PACKAGE_VERSION": "1.5.1",
+    "PACKAGE_VERSION_MAJOR": "1",
+    "PACKAGE_VERSION_MICRO": "1",
+    "PACKAGE_VERSION_MINOR": "5",
+    "SHORT_LIBZIP": 2,
+    "SIZEOF_OFF_T": 8,
+    "SIZE_T_LIBZIP": 8,
+    "SSIZE_T_LIBZIP": 8,
+    "UINT16_T_LIBZIP": 2,
+    "UINT32_T_LIBZIP": 4,
+    "UINT64_T_LIBZIP": 8,
+    "UINT8_T_LIBZIP": 1,
+    "__INT16_LIBZIP": None,
+    "__INT32_LIBZIP": None,
+    "__INT64_LIBZIP": None,
+    "__INT8_LIBZIP": None,
+}
+
+_CMAKE_VARIABLES.update(dict([
+    (
+        "ZIP_{sign}INT{size}_T".format(
+            sign = sign.upper(),
+            size = size,
+        ),
+        "{sign}int{size}_t".format(
+            sign = sign.lower(),
+            size = size,
+        ),
+    )
+    for sign in ("U", "")
+    for size in (8, 16, 32, 64)
+]))
+
+_SUBSTITUTIONS = {
+    "@PACKAGE@": "libzip",
+    "@VERSION@": "1.5.1",  # Keep in sync with actual package!
+}
+
+_DEFINES = {
+    "HAVE_CLONEFILE": False,
+    "HAVE_COMMONCRYPTO": False,
+    "HAVE_CRYPTO": False,
+    "HAVE_DIRENT_H": False,
+    "HAVE_FICLONERANGE": False,
+    "HAVE_FILENO": True,
+    "HAVE_FSEEK": True,
+    "HAVE_FSEEKO": True,
+    "HAVE_FTELLO": True,
+    "HAVE_FTS_H": True,
+    "HAVE_GETPROGNAME": False,
+    "HAVE_GNUTLS": False,
+    "HAVE_LIBBZ2": False,
+    "HAVE_MKSTEMP": True,
+    "HAVE_NDIR_H": False,
+    "HAVE_OPEN": True,
+    "HAVE_OPENSSL": False,
+    "HAVE_SETMODE": False,
+    "HAVE_SHARED": True,
+    "HAVE_SNPRINTF": True,
+    "HAVE_SSIZE_T_LIBZIP": True,
+    "HAVE_STDBOOL_H": True,
+    "HAVE_STRCASECMP": True,
+    "HAVE_STRDUP": True,
+    "HAVE_STRICMP": False,
+    "HAVE_STRINGS_H": True,
+    "HAVE_STRTOLL": True,
+    "HAVE_STRTOULL": True,
+    "HAVE_STRUCT_TM_TM_ZONE": False,
+    "HAVE_SYS_DIR_H": False,
+    "HAVE_SYS_NDIR_H": False,
+    "HAVE_UNISTD_H": True,
+    "HAVE__CHMOD": False,
+    "HAVE__CLOSE": False,
+    "HAVE__DUP": False,
+    "HAVE__FDOPEN": False,
+    "HAVE__FILENO": False,
+    "HAVE__OPEN": False,
+    "HAVE__SETMODE": False,
+    "HAVE__SNPRINTF": False,
+    "HAVE__STRDUP": False,
+    "HAVE__STRICMP": False,
+    "HAVE__STRTOI64": False,
+    "HAVE__STRTOUI64": False,
+    "HAVE__UMASK": False,
+    "HAVE__UNLINK": False,
+    "HAVE___PROGNAME": False,
+    "WORDS_BIGENDIAN": False,
+}
+
+_DEFINES.update(dict([(
+    key,
+    value != None,
+) for key, value in _CMAKE_VARIABLES.items()]))
+
+_SUBSTITUTIONS.update(cmake_substitutions(
+    defines = _DEFINES,
+    vars = _CMAKE_VARIABLES,
+))
+
+expand_template(
+    name = "config_h",
+    out = "config.h",
+    substitutions = _SUBSTITUTIONS,
+    template = "cmake-config.h.in",
+)
+
+_VARS = {
+    "LIBZIP_TYPES_INCLUDE": "#include <stdint.h>",
+    "PACKAGE_VERSION": "1.5.1",
+    "PACKAGE_VERSION_MAJOR": "1",
+    "PACKAGE_VERSION_MICRO": "1",
+    "PACKAGE_VERSION_MINOR": "5",
+}
+
+_VARS.update(dict([
+    (
+        "ZIP_{sign}INT{size}_T".format(
+            sign = sign.upper(),
+            size = size,
+        ),
+        "{sign}int{size}_t".format(
+            sign = sign.lower(),
+            size = size,
+        ),
+    )
+    for sign in ("U", "")
+    for size in (8, 16, 32, 64)
+]))
+
+expand_template(
+    name = "zipconf_h",
+    out = "lib/zipconf.h",
+    substitutions = cmake_substitutions(
+        defines = {
+            "LIBZIP_VERSION": True,
+            "LIBZIP_VERSION_MAJOR": True,
+            "LIBZIP_VERSION_MICRO": True,
+            "LIBZIP_VERSION_MINOR": True,
+            "ZIP_STATIC": False,
+        },
+        vars = _VARS,
+    ),
+    template = "cmake-zipconf.h.in",
+)
+
+cc_library(
+    name = "zip",
+    srcs = glob(
+        [
+            "lib/*.c",
+            "lib/*.h",
+        ],
+        exclude = [
+            "lib/*win32*",
+            "lib/zip_random_uwp.c",
+            "lib/*crypto*",
+            "lib/*aes*",
+            "lib/*bzip2*",
+        ],
+    ) + [
+        "config.h",
+    ],
+    hdrs = [
+        "lib/zip.h",
+        "lib/zipconf.h",
+    ],
+    copts = [
+        "-DHAVE_CONFIG_H",
+    ],
+    includes = ["lib"],
+    deps = [
+        "@zlib",
+    ],
+)
diff --git a/third_party/py/BUILD b/third_party/py/BUILD
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/third_party/py/BUILD
diff --git a/third_party/py/BUILD.tpl b/third_party/py/BUILD.tpl
new file mode 100644
index 0000000..cc0e013
--- /dev/null
+++ b/third_party/py/BUILD.tpl
@@ -0,0 +1,31 @@
+licenses(["restricted"])
+
+package(default_visibility = ["//visibility:public"])
+
+# Point both runtimes to the same python binary to ensure we always
+# use the python binary specified by ./configure.py script.
+load("@bazel_tools//tools/python:toolchain.bzl", "py_runtime_pair")
+
+py_runtime(
+    name = "py2_runtime",
+    interpreter_path = "%{PYTHON_BIN_PATH}",
+    python_version = "PY2",
+)
+
+py_runtime(
+    name = "py3_runtime",
+    interpreter_path = "%{PYTHON_BIN_PATH}",
+    python_version = "PY3",
+)
+
+py_runtime_pair(
+    name = "py_runtime_pair",
+    py2_runtime = ":py2_runtime",
+    py3_runtime = ":py3_runtime",
+)
+
+toolchain(
+    name = "py_toolchain",
+    toolchain = ":py_runtime_pair",
+    toolchain_type = "@bazel_tools//tools/python:toolchain_type",
+)
diff --git a/third_party/py/python_configure.bzl b/third_party/py/python_configure.bzl
new file mode 100644
index 0000000..6601d7f
--- /dev/null
+++ b/third_party/py/python_configure.bzl
@@ -0,0 +1,71 @@
+"""Repository rule for Python autoconfiguration.
+
+`python_configure` depends on the following environment variables:
+
+  * `PYTHON_BIN_PATH`: location of python binary.
+"""
+
+_PYTHON_BIN_PATH = "PYTHON_BIN_PATH"
+
+def _tpl(repository_ctx, tpl, substitutions = {}, out = None):
+    if not out:
+        out = tpl
+    repository_ctx.template(
+        out,
+        Label("//third_party/py:%s.tpl" % tpl),
+        substitutions,
+    )
+
+def _fail(msg):
+    """Output failure message when auto configuration fails."""
+    red = "\033[0;31m"
+    no_color = "\033[0m"
+    fail("%sPython Configuration Error:%s %s\n" % (red, no_color, msg))
+
+def _get_python_bin(repository_ctx):
+    """Gets the python bin path."""
+    python_bin = repository_ctx.os.environ.get(_PYTHON_BIN_PATH)
+    if python_bin != None:
+        return python_bin
+    python_bin_path = repository_ctx.which("python")
+    if python_bin_path != None:
+        return str(python_bin_path)
+    _fail("Cannot find python in PATH, please make sure " +
+          "python is installed and add its directory in PATH, or --define " +
+          "%s='/something/else'.\nPATH=%s" % (
+              _PYTHON_BIN_PATH,
+              repository_ctx.os.environ.get("PATH", ""),
+          ))
+
+def _create_local_python_repository(repository_ctx):
+    """Creates the repository containing files set up to build with Python."""
+    python_bin = _get_python_bin(repository_ctx)
+    _tpl(repository_ctx, "BUILD", {
+        "%{PYTHON_BIN_PATH}": python_bin,
+    })
+
+def _python_autoconf_impl(repository_ctx):
+    """Implementation of the python_autoconf repository rule."""
+    _create_local_python_repository(repository_ctx)
+
+python_configure = repository_rule(
+    implementation = _python_autoconf_impl,
+    environ = [
+        _PYTHON_BIN_PATH,
+    ],
+)
+"""Detects and configures the local Python toolchain.
+
+Add the following to your WORKSPACE FILE:
+
+```python
+load("//third_party/py:python_configure.bzl", "python_configure")
+
+python_configure(name = "local_config_py_toolchain")
+
+register_toolchains("@local_config_py_toolchain//:py_toolchain")
+```
+
+Args:
+  name: A unique name for this workspace rule.
+"""
diff --git a/third_party/pybind11.BUILD b/third_party/pybind11.BUILD
new file mode 100644
index 0000000..2f1ada6
--- /dev/null
+++ b/third_party/pybind11.BUILD
@@ -0,0 +1,25 @@
+package(default_visibility = ["//visibility:public"])
+
+cc_library(
+    name = "pybind11",
+    hdrs = glob(
+        include = [
+            "include/pybind11/*.h",
+            "include/pybind11/detail/*.h",
+        ],
+        exclude = [
+            "include/pybind11/common.h",
+            "include/pybind11/eigen.h",
+        ],
+    ),
+    copts = [
+        "-fexceptions",
+        "-Wno-undefined-inline",
+        "-Wno-pragma-once-outside-header",
+    ],
+    includes = ["include"],
+    strip_include_prefix = "include",
+    deps = [
+        "@org_tensorflow//third_party/python_runtime:headers",
+    ],
+)
diff --git a/third_party/python_runtime/BUILD b/third_party/python_runtime/BUILD
new file mode 100644
index 0000000..2a16091
--- /dev/null
+++ b/third_party/python_runtime/BUILD
@@ -0,0 +1,8 @@
+licenses(["notice"])  # New BSD, Python Software Foundation
+
+package(default_visibility = ["//visibility:public"])
+
+alias(
+    name = "headers",
+    actual = "@local_config_python//:python_headers",
+)
diff --git a/third_party/repo.bzl b/third_party/repo.bzl
new file mode 100644
index 0000000..c9c6a83
--- /dev/null
+++ b/third_party/repo.bzl
@@ -0,0 +1,152 @@
+# Copyright 2020 The TensorFlow Authors. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Utilities for defining TensorFlow Lite Support Bazel dependencies."""
+
+_SINGLE_URL_WHITELIST = []
+
+def _is_windows(ctx):
+    return ctx.os.name.lower().find("windows") != -1
+
+def _wrap_bash_cmd(ctx, cmd):
+    if _is_windows(ctx):
+        bazel_sh = _get_env_var(ctx, "BAZEL_SH")
+        if not bazel_sh:
+            fail("BAZEL_SH environment variable is not set")
+        cmd = [bazel_sh, "-l", "-c", " ".join(["\"%s\"" % s for s in cmd])]
+    return cmd
+
+def _get_env_var(ctx, name):
+    if name in ctx.os.environ:
+        return ctx.os.environ[name]
+    else:
+        return None
+
+# Checks if we should use the system lib instead of the bundled one
+def _use_system_lib(ctx, name):
+    syslibenv = _get_env_var(ctx, "TF_SYSTEM_LIBS")
+    if syslibenv:
+        for n in syslibenv.strip().split(","):
+            if n.strip() == name:
+                return True
+    return False
+
+# Executes specified command with arguments and calls 'fail' if it exited with
+# non-zero code
+def _execute_and_check_ret_code(repo_ctx, cmd_and_args):
+    result = repo_ctx.execute(cmd_and_args, timeout = 60)
+    if result.return_code != 0:
+        fail(("Non-zero return code({1}) when executing '{0}':\n" + "Stdout: {2}\n" +
+              "Stderr: {3}").format(
+            " ".join([str(x) for x in cmd_and_args]),
+            result.return_code,
+            result.stdout,
+            result.stderr,
+        ))
+
+# Apply a patch_file to the repository root directory
+# Runs 'patch -p1' on both Windows and Unix.
+def _apply_patch(ctx, patch_file):
+    patch_command = ["patch", "-p1", "-d", ctx.path("."), "-i", ctx.path(patch_file)]
+    cmd = _wrap_bash_cmd(ctx, patch_command)
+    _execute_and_check_ret_code(ctx, cmd)
+
+def _apply_delete(ctx, paths):
+    for path in paths:
+        if path.startswith("/"):
+            fail("refusing to rm -rf path starting with '/': " + path)
+        if ".." in path:
+            fail("refusing to rm -rf path containing '..': " + path)
+    cmd = _wrap_bash_cmd(ctx, ["rm", "-rf"] + [ctx.path(path) for path in paths])
+    _execute_and_check_ret_code(ctx, cmd)
+
+def _third_party_http_archive(ctx):
+    """Downloads and creates Bazel repos for dependencies.
+
+    This is a swappable replacement for both http_archive() and
+    new_http_archive() that offers some additional features. It also helps
+    ensure best practices are followed.
+    """
+    if ("mirror.tensorflow.org" not in ctx.attr.urls[0] and
+        (len(ctx.attr.urls) < 2 and
+         ctx.attr.name not in _SINGLE_URL_WHITELIST.to_list())):
+        fail("third_party_http_archive(urls) must have redundant URLs. The " +
+             "mirror.tensorflow.org URL must be present and it must come first. " +
+             "Even if you don't have permission to mirror the file, please " +
+             "put the correctly formatted mirror URL there anyway, because " +
+             "someone will come along shortly thereafter and mirror the file.")
+
+    use_syslib = _use_system_lib(ctx, ctx.attr.name)
+
+    # Use "BUILD.bazel" to avoid conflict with third party projects that contain a
+    # file or directory called "BUILD"
+    buildfile_path = ctx.path("BUILD.bazel")
+
+    if use_syslib:
+        if ctx.attr.system_build_file == None:
+            fail("Bazel was configured with TF_SYSTEM_LIBS to use a system " +
+                 "library for %s, but no system build file for %s was configured. " +
+                 "Please add a system_build_file attribute to the repository rule" +
+                 "for %s." % (ctx.attr.name, ctx.attr.name, ctx.attr.name))
+        ctx.symlink(Label(ctx.attr.system_build_file), buildfile_path)
+
+    else:
+        ctx.download_and_extract(
+            ctx.attr.urls,
+            "",
+            ctx.attr.sha256,
+            ctx.attr.type,
+            ctx.attr.strip_prefix,
+        )
+        if ctx.attr.delete:
+            _apply_delete(ctx, ctx.attr.delete)
+        if ctx.attr.patch_file != None:
+            _apply_patch(ctx, ctx.attr.patch_file)
+        ctx.symlink(Label(ctx.attr.build_file), buildfile_path)
+
+    link_dict = {}
+    if use_syslib:
+        link_dict.update(ctx.attr.system_link_files)
+
+    for internal_src, external_dest in ctx.attr.link_files.items():
+        # if syslib and link exists in both, use the system one
+        if external_dest not in link_dict.values():
+            link_dict[internal_src] = external_dest
+
+    for internal_src, external_dest in link_dict.items():
+        ctx.symlink(Label(internal_src), ctx.path(external_dest))
+
+# For link_files, specify each dict entry as:
+# "//path/to/source:file": "localfile"
+third_party_http_archive = repository_rule(
+    attrs = {
+        "sha256": attr.string(mandatory = True),
+        "urls": attr.string_list(
+            mandatory = True,
+            allow_empty = False,
+        ),
+        "strip_prefix": attr.string(),
+        "type": attr.string(),
+        "delete": attr.string_list(),
+        "build_file": attr.string(mandatory = True),
+        "system_build_file": attr.string(mandatory = False),
+        "patch_file": attr.label(),
+        "link_files": attr.string_dict(),
+        "system_link_files": attr.string_dict(),
+    },
+    environ = [
+        "TF_SYSTEM_LIBS",
+    ],
+    implementation = _third_party_http_archive,
+)
diff --git a/third_party/six.BUILD b/third_party/six.BUILD
new file mode 100644
index 0000000..a1b2f7b
--- /dev/null
+++ b/third_party/six.BUILD
@@ -0,0 +1,14 @@
+# Description:
+#   Six provides simple utilities for wrapping over differences between Python 2
+#   and Python 3.
+
+licenses(["notice"])  # MIT
+
+exports_files(["LICENSE"])
+
+py_library(
+    name = "six",
+    srcs = ["six.py"],
+    srcs_version = "PY2AND3",
+    visibility = ["//visibility:public"],
+)
diff --git a/third_party/stblib.BUILD b/third_party/stblib.BUILD
new file mode 100644
index 0000000..f1c361a
--- /dev/null
+++ b/third_party/stblib.BUILD
@@ -0,0 +1,26 @@
+# Description:
+#   Single-file C++ image decoding and encoding libraries
+
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+licenses(["notice"])  # MIT license
+
+exports_files(["LICENSE"])
+
+cc_library(
+    name = "stb_image",
+    hdrs = ["stb_image.h"],
+    copts = [
+        "-Wno-unused-function",
+        "$(STACK_FRAME_UNLIMITED)",
+    ],
+    includes = ["."],
+)
+
+cc_library(
+    name = "stb_image_write",
+    hdrs = ["stb_image_write.h"],
+    includes = ["."],
+)
diff --git a/third_party/tensorflow/BUILD b/third_party/tensorflow/BUILD
new file mode 100644
index 0000000..ac039c4
--- /dev/null
+++ b/third_party/tensorflow/BUILD
@@ -0,0 +1 @@
+# placeholder to make the directory a bazel package.
diff --git a/third_party/tensorflow/BUILD.tpl b/third_party/tensorflow/BUILD.tpl
new file mode 100644
index 0000000..095021e
--- /dev/null
+++ b/third_party/tensorflow/BUILD.tpl
@@ -0,0 +1,18 @@
+package(default_visibility = ["//visibility:public"])
+
+cc_library(
+    name = "tf_header_lib",
+    hdrs = [":tf_header_include"],
+    includes = ["include"],
+    visibility = ["//visibility:public"],
+)
+
+cc_library(
+    name = "libtensorflow_framework",
+    srcs = [":libtensorflow_framework.so"],
+    visibility = ["//visibility:public"],
+)
+
+%{TF_HEADER_GENRULE}
+%{TF_SHARED_LIBRARY_GENRULE}
+
diff --git a/third_party/tensorflow/tf_configure.bzl b/third_party/tensorflow/tf_configure.bzl
new file mode 100644
index 0000000..3282625
--- /dev/null
+++ b/third_party/tensorflow/tf_configure.bzl
@@ -0,0 +1,224 @@
+"""Setup TensorFlow as external dependency"""
+
+_TF_HEADER_DIR = "TF_HEADER_DIR"
+_TF_SHARED_LIBRARY_DIR = "TF_SHARED_LIBRARY_DIR"
+_TF_SHARED_LIBRARY_NAME = "TF_SHARED_LIBRARY_NAME"
+
+def _tpl(repository_ctx, tpl, substitutions = {}, out = None):
+    if not out:
+        out = tpl
+    repository_ctx.template(
+        out,
+        Label("//third_party/tensorflow:%s.tpl" % tpl),
+        substitutions,
+    )
+
+def _fail(msg):
+    """Output failure message when auto configuration fails."""
+    red = "\033[0;31m"
+    no_color = "\033[0m"
+    fail("%sPython Configuration Error:%s %s\n" % (red, no_color, msg))
+
+def _is_windows(repository_ctx):
+    """Returns true if the host operating system is windows."""
+    os_name = repository_ctx.os.name.lower()
+    if os_name.find("windows") != -1:
+        return True
+    return False
+
+def _execute(
+        repository_ctx,
+        cmdline,
+        error_msg = None,
+        error_details = None,
+        empty_stdout_fine = False):
+    """Executes an arbitrary shell command.
+
+    Helper for executes an arbitrary shell command.
+
+    Args:
+      repository_ctx: the repository_ctx object.
+      cmdline: list of strings, the command to execute.
+      error_msg: string, a summary of the error if the command fails.
+      error_details: string, details about the error or steps to fix it.
+      empty_stdout_fine: bool, if True, an empty stdout result is fine, otherwise
+        it's an error.
+
+    Returns:
+      The result of repository_ctx.execute(cmdline).
+    """
+    result = repository_ctx.execute(cmdline)
+    if result.stderr or not (empty_stdout_fine or result.stdout):
+        _fail("\n".join([
+            error_msg.strip() if error_msg else "Repository command failed",
+            result.stderr.strip(),
+            error_details if error_details else "",
+        ]))
+    return result
+
+def _read_dir(repository_ctx, src_dir):
+    """Returns a string with all files in a directory.
+
+    Finds all files inside a directory, traversing subfolders and following
+    symlinks. The returned string contains the full path of all files
+    separated by line breaks.
+
+    Args:
+        repository_ctx: the repository_ctx object.
+        src_dir: directory to find files from.
+
+    Returns:
+        A string of all files inside the given dir.
+    """
+    if _is_windows(repository_ctx):
+        src_dir = src_dir.replace("/", "\\")
+        find_result = _execute(
+            repository_ctx,
+            ["cmd.exe", "/c", "dir", src_dir, "/b", "/s", "/a-d"],
+            empty_stdout_fine = True,
+        )
+
+        # src_files will be used in genrule.outs where the paths must
+        # use forward slashes.
+        result = find_result.stdout.replace("\\", "/")
+    else:
+        find_result = _execute(
+            repository_ctx,
+            ["find", src_dir, "-follow", "-type", "f"],
+            empty_stdout_fine = True,
+        )
+        result = find_result.stdout
+    return result
+
+def _genrule(genrule_name, command, outs):
+    """Returns a string with a genrule.
+
+    Genrule executes the given command and produces the given outputs.
+
+    Args:
+        genrule_name: A unique name for genrule target.
+        command: The command to run.
+        outs: A list of files generated by this rule.
+
+    Returns:
+        A genrule target.
+    """
+    return (
+        "genrule(\n" +
+        '    name = "' +
+        genrule_name + '",\n' +
+        "    outs = [\n" +
+        outs +
+        "\n    ],\n" +
+        '    cmd = """\n' +
+        command +
+        '\n   """,\n' +
+        ")\n"
+    )
+
+def _norm_path(path):
+    """Returns a path with '/' and remove the trailing slash."""
+    path = path.replace("\\", "/")
+    if path[-1] == "/":
+        path = path[:-1]
+    return path
+
+def _symlink_genrule_for_dir(
+        repository_ctx,
+        src_dir,
+        dest_dir,
+        genrule_name,
+        src_files = [],
+        dest_files = [],
+        tf_pip_dir_rename_pair = []):
+    """Returns a genrule to symlink(or copy if on Windows) a set of files.
+
+    If src_dir is passed, files will be read from the given directory; otherwise
+    we assume files are in src_files and dest_files.
+
+    Args:
+        repository_ctx: the repository_ctx object.
+        src_dir: source directory.
+        dest_dir: directory to create symlink in.
+        genrule_name: genrule name.
+        src_files: list of source files instead of src_dir.
+        dest_files: list of corresonding destination files.
+        tf_pip_dir_rename_pair: list of the pair of tf pip parent directory to
+          replace. For example, in TF pip package, the source code is under
+          "tensorflow_core", and we might want to replace it with
+          "tensorflow" to match the header includes.
+    Returns:
+        genrule target that creates the symlinks.
+    """
+
+    # Check that tf_pip_dir_rename_pair has the right length
+    tf_pip_dir_rename_pair_len = len(tf_pip_dir_rename_pair)
+    if tf_pip_dir_rename_pair_len != 0 and tf_pip_dir_rename_pair_len != 2:
+        _fail("The size of argument tf_pip_dir_rename_pair should be either 0 or 2, but %d is given." % tf_pip_dir_rename_pair_len)
+
+    if src_dir != None:
+        src_dir = _norm_path(src_dir)
+        dest_dir = _norm_path(dest_dir)
+        files = "\n".join(sorted(_read_dir(repository_ctx, src_dir).splitlines()))
+
+        # Create a list with the src_dir stripped to use for outputs.
+        if tf_pip_dir_rename_pair_len:
+            dest_files = files.replace(src_dir, "").replace(tf_pip_dir_rename_pair[0], tf_pip_dir_rename_pair[1]).splitlines()
+        else:
+            dest_files = files.replace(src_dir, "").splitlines()
+        src_files = files.splitlines()
+    command = []
+    outs = []
+    for i in range(len(dest_files)):
+        if dest_files[i] != "":
+            # If we have only one file to link we do not want to use the dest_dir, as
+            # $(@D) will include the full path to the file.
+            dest = "$(@D)/" + dest_dir + dest_files[i] if len(dest_files) != 1 else "$(@D)/" + dest_files[i]
+
+            # Copy the headers to create a sandboxable setup.
+            cmd = "cp -f"
+            command.append(cmd + ' "%s" "%s"' % (src_files[i], dest))
+            outs.append('        "' + dest_dir + dest_files[i] + '",')
+    dest_dir = "abc"
+    genrule = _genrule(
+        genrule_name,
+        " && ".join(command),
+        "\n".join(outs),
+    )
+    return genrule
+
+def _tf_pip_impl(repository_ctx):
+    tf_header_dir = repository_ctx.os.environ[_TF_HEADER_DIR]
+    tf_header_rule = _symlink_genrule_for_dir(
+        repository_ctx,
+        tf_header_dir,
+        "include",
+        "tf_header_include",
+        tf_pip_dir_rename_pair = ["tensorflow_core", "tensorflow"],
+    )
+
+    tf_shared_library_dir = repository_ctx.os.environ[_TF_SHARED_LIBRARY_DIR]
+    tf_shared_library_name = repository_ctx.os.environ[_TF_SHARED_LIBRARY_NAME]
+    tf_shared_library_path = "%s/%s" % (tf_shared_library_dir, tf_shared_library_name)
+    tf_shared_library_rule = _symlink_genrule_for_dir(
+        repository_ctx,
+        None,
+        "",
+        "libtensorflow_framework.so",
+        [tf_shared_library_path],
+        ["_pywrap_tensorflow_internal.lib" if _is_windows(repository_ctx) else "libtensorflow_framework.so"],
+    )
+
+    _tpl(repository_ctx, "BUILD", {
+        "%{TF_HEADER_GENRULE}": tf_header_rule,
+        "%{TF_SHARED_LIBRARY_GENRULE}": tf_shared_library_rule,
+    })
+
+tf_configure = repository_rule(
+    implementation = _tf_pip_impl,
+    environ = [
+        _TF_HEADER_DIR,
+        _TF_SHARED_LIBRARY_DIR,
+        _TF_SHARED_LIBRARY_NAME,
+    ],
+)
diff --git a/third_party/tensorflow_lite_ios_build.patch b/third_party/tensorflow_lite_ios_build.patch
new file mode 100644
index 0000000..786e46b
--- /dev/null
+++ b/third_party/tensorflow_lite_ios_build.patch
@@ -0,0 +1,40 @@
+diff --git a/tensorflow/lite/experimental/ios/BUILD.apple b/tensorflow/lite/experimental/ios/BUILD
+similarity index 97%
+rename from tensorflow/lite/experimental/ios/BUILD.apple
+rename to tensorflow/lite/experimental/ios/BUILD
+index cce0c4df..49eba35f 100644
+--- a/tensorflow/lite/experimental/ios/BUILD.apple
++++ b/tensorflow/lite/experimental/ios/BUILD
+@@ -22,8 +22,7 @@ sh_binary(
+         "hide_symbols_with_allowlist.sh",
+     ],
+     visibility = [
+-        "//tensorflow/lite:__subpackages__",
+-        "//tensorflow_lite_support:__subpackages__",
++        "//visibility:public",
+     ],
+ )
+ 
+diff --git a/tensorflow/lite/experimental/ios/ios.bzl b/tensorflow/lite/experimental/ios/ios.bzl
+index 63747eb8..07bcb49d 100644
+--- a/tensorflow/lite/experimental/ios/ios.bzl
++++ b/tensorflow/lite/experimental/ios/ios.bzl
+@@ -60,7 +60,7 @@ def tflite_ios_static_framework(
+            "BUNDLE_NAME=\"" + bundle_name + "\" " +
+            "ALLOWLIST_FILE_PATH=\"$(location " + allowlist_symbols_file + ")\" " +
+            "OUTPUT=\"$(OUTS)\" " +
+-           "\"$(location //tensorflow/lite/experimental/ios:hide_symbols_with_allowlist)\"")
++           "\"$(location @org_tensorflow//tensorflow/lite/experimental/ios:hide_symbols_with_allowlist)\"")
+ 
+     native.genrule(
+         name = name,
+@@ -68,7 +68,7 @@ def tflite_ios_static_framework(
+         outs = [name + ".zip"],
+         cmd = cmd,
+         tools = [
+-            "//tensorflow/lite/experimental/ios:hide_symbols_with_allowlist",
++            "@org_tensorflow//tensorflow/lite/experimental/ios:hide_symbols_with_allowlist",
+         ],
+     )
+ 
+
diff --git a/third_party/tensorflow_text_remove_tf_deps.patch b/third_party/tensorflow_text_remove_tf_deps.patch
new file mode 100644
index 0000000..f7b86f9
--- /dev/null
+++ b/third_party/tensorflow_text_remove_tf_deps.patch
@@ -0,0 +1,32 @@
+diff --git a/tensorflow_text/core/kernels/BUILD b/tensorflow_text/core/kernels/BUILD
+index bdca365..1c20eae 100644
+--- a/tensorflow_text/core/kernels/BUILD
++++ b/tensorflow_text/core/kernels/BUILD
+@@ -209,8 +209,12 @@ cc_library(
+     name = "regex_split",
+     srcs = ["regex_split.cc"],
+     hdrs = ["regex_split.h"],
+-    deps = OSS_DEPS + [
++    deps = [
+         # absl/strings dep
++        "@com_google_absl//absl/container:inlined_vector",
++        "@com_google_absl//absl/strings",
++        "@com_google_absl//absl/types:optional",
++        "@com_google_absl//absl/types:span",
+         "@com_google_re2//:re2",
+     ],
+ )
+@@ -437,8 +441,12 @@ cc_library(
+     name = "wordpiece_tokenizer",
+     srcs = ["wordpiece_tokenizer.cc"],
+     hdrs = ["wordpiece_tokenizer.h"],
+-    deps = OSS_DEPS + [
++    deps = [
+         # absl/strings dep
++        "@com_google_absl//absl/container:inlined_vector",
++        "@com_google_absl//absl/strings",
++        "@com_google_absl//absl/types:optional",
++        "@com_google_absl//absl/types:span",
+         "@icu//:common",
+     ],
+ )
\ No newline at end of file
diff --git a/third_party/toolchains/java/BUILD b/third_party/toolchains/java/BUILD
new file mode 100644
index 0000000..8372291
--- /dev/null
+++ b/third_party/toolchains/java/BUILD
@@ -0,0 +1,18 @@
+# For workaround https://github.com/bazelbuild/bazel/issues/8772 with Bazel >= 0.29.1
+# TensorFlow still targets Java 1.7 (See JAVACOPTS in tensorflow/java/build_defs.bzl)
+# which doesn't support "-parameters" flag. Starting from Java 11 (default since Bazel
+# 0.29.1), a warning message will be thrown if "-parameters" is passed. If "-Werror" also exists,
+# the compiling action will fail. To workaround this, we override the misc value of
+# the default java toolchain to remove "-parameters" flag.
+load("@bazel_tools//tools/jdk:default_java_toolchain.bzl", "default_java_toolchain")
+
+licenses(["notice"])
+
+default_java_toolchain(
+    name = "tf_java_toolchain",
+    misc = [
+        "-XDskipDuplicateBridges=true",
+        "-g",
+    ],
+    visibility = ["//visibility:public"],
+)
diff --git a/third_party/utf.BUILD b/third_party/utf.BUILD
new file mode 100644
index 0000000..0a21fd7
--- /dev/null
+++ b/third_party/utf.BUILD
@@ -0,0 +1,38 @@
+cc_library(
+    name = "utf",
+    srcs = [
+        "libutf/rune.c",
+        "libutf/runestrcat.c",
+        "libutf/runestrchr.c",
+        "libutf/runestrcmp.c",
+        "libutf/runestrcpy.c",
+        "libutf/runestrdup.c",
+        "libutf/runestrecpy.c",
+        "libutf/runestrlen.c",
+        "libutf/runestrncat.c",
+        "libutf/runestrncmp.c",
+        "libutf/runestrncpy.c",
+        "libutf/runestrrchr.c",
+        "libutf/runestrstr.c",
+        "libutf/runetype.c",
+        "libutf/utfecpy.c",
+        "libutf/utflen.c",
+        "libutf/utfnlen.c",
+        "libutf/utfrrune.c",
+        "libutf/utfrune.c",
+        "libutf/utfutf.c",
+    ],
+    hdrs = [
+        "libutf/plan9.h",
+        "libutf/utf.h",
+        "libutf/utfdef.h",
+    ],
+    copts = [
+        "-Wno-parentheses",
+    ],
+    includes = [
+        ".",
+        "libutf",
+    ],
+    visibility = ["//visibility:public"],
+)
diff --git a/third_party/zlib.BUILD b/third_party/zlib.BUILD
new file mode 100644
index 0000000..275782e
--- /dev/null
+++ b/third_party/zlib.BUILD
@@ -0,0 +1,39 @@
+package(default_visibility = ["//visibility:public"])
+
+licenses(["notice"])
+
+cc_library(
+    name = "zlib",
+    srcs = [
+        "adler32.c",
+        "compress.c",
+        "crc32.c",
+        "crc32.h",
+        "deflate.c",
+        "deflate.h",
+        "gzclose.c",
+        "gzguts.h",
+        "gzlib.c",
+        "gzread.c",
+        "gzwrite.c",
+        "infback.c",
+        "inffast.c",
+        "inffast.h",
+        "inffixed.h",
+        "inflate.c",
+        "inflate.h",
+        "inftrees.c",
+        "inftrees.h",
+        "trees.c",
+        "trees.h",
+        "uncompr.c",
+        "zutil.c",
+        "zutil.h",
+    ],
+    hdrs = [
+        "zconf.h",
+        "zlib.h",
+    ],
+    copts = ["-Wno-implicit-function-declaration"],
+    includes = ["."],
+)