tensorflow_lite_support/cc/task/vision/proto/segmentations.proto

/* 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;
}