blob: eae2167574f0a274ab0755e5310096aa1943c2f8 [file] [log] [blame]
<!-- HTML header for doxygen 1.8.9.1-->
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
<meta http-equiv="X-UA-Compatible" content="IE=9"/>
<meta name="generator" content="Doxygen 1.8.6"/>
<meta name="robots" content="NOINDEX, NOFOLLOW" /> <!-- Prevent indexing by search engines -->
<title>Compute Library: Importing data from existing models</title>
<link href="tabs.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="jquery.js"></script>
<script type="text/javascript" src="dynsections.js"></script>
<link href="navtree.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="resize.js"></script>
<script type="text/javascript" src="navtree.js"></script>
<script type="text/javascript">
$(document).ready(initResizable);
$(window).load(resizeHeight);
</script>
<link href="search/search.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="search/search.js"></script>
<script type="text/javascript">
$(document).ready(function() { searchBox.OnSelectItem(0); });
</script>
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
extensions: ["tex2jax.js"],
jax: ["input/TeX","output/HTML-CSS"],
});
</script><script src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script>
<link href="doxygen.css" rel="stylesheet" type="text/css" />
</head>
<body>
<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
<div id="titlearea">
<table cellspacing="0" cellpadding="0">
<tbody>
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">Compute Library
&#160;<span id="projectnumber">17.09</span>
</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<!-- Generated by Doxygen 1.8.6 -->
<script type="text/javascript">
var searchBox = new SearchBox("searchBox", "search",false,'Search');
</script>
<div id="navrow1" class="tabs">
<ul class="tablist">
<li><a href="index.xhtml"><span>Main&#160;Page</span></a></li>
<li class="current"><a href="pages.xhtml"><span>Related&#160;Pages</span></a></li>
<li><a href="namespaces.xhtml"><span>Namespaces</span></a></li>
<li><a href="annotated.xhtml"><span>Data&#160;Structures</span></a></li>
<li><a href="files.xhtml"><span>Files</span></a></li>
<li>
<div id="MSearchBox" class="MSearchBoxInactive">
<span class="left">
<img id="MSearchSelect" src="search/mag_sel.png"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
alt=""/>
<input type="text" id="MSearchField" value="Search" accesskey="S"
onfocus="searchBox.OnSearchFieldFocus(true)"
onblur="searchBox.OnSearchFieldFocus(false)"
onkeyup="searchBox.OnSearchFieldChange(event)"/>
</span><span class="right">
<a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a>
</span>
</div>
</li>
</ul>
</div>
</div><!-- top -->
<div id="side-nav" class="ui-resizable side-nav-resizable">
<div id="nav-tree">
<div id="nav-tree-contents">
<div id="nav-sync" class="sync"></div>
</div>
</div>
<div id="splitbar" style="-moz-user-select:none;"
class="ui-resizable-handle">
</div>
</div>
<script type="text/javascript">
$(document).ready(function(){initNavTree('data_import.xhtml','');});
</script>
<div id="doc-content">
<!-- window showing the filter options -->
<div id="MSearchSelectWindow"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
onkeydown="return searchBox.OnSearchSelectKey(event)">
<a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(0)"><span class="SelectionMark">&#160;</span>All</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(1)"><span class="SelectionMark">&#160;</span>Data Structures</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(2)"><span class="SelectionMark">&#160;</span>Namespaces</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(3)"><span class="SelectionMark">&#160;</span>Files</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(4)"><span class="SelectionMark">&#160;</span>Functions</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(5)"><span class="SelectionMark">&#160;</span>Variables</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(6)"><span class="SelectionMark">&#160;</span>Typedefs</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(7)"><span class="SelectionMark">&#160;</span>Enumerations</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(8)"><span class="SelectionMark">&#160;</span>Enumerator</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(9)"><span class="SelectionMark">&#160;</span>Friends</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(10)"><span class="SelectionMark">&#160;</span>Macros</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(11)"><span class="SelectionMark">&#160;</span>Pages</a></div>
<!-- iframe showing the search results (closed by default) -->
<div id="MSearchResultsWindow">
<iframe src="javascript:void(0)" frameborder="0"
name="MSearchResults" id="MSearchResults">
</iframe>
</div>
<div class="header">
<div class="headertitle">
<div class="title">Importing data from existing models </div> </div>
</div><!--header-->
<div class="contents">
<div class="toc"><h3>Table of Contents</h3>
<ul><li class="level1"><a href="#caffe_data_extractor">Extract data from pre-trained caffe model</a><ul><li class="level2"><a href="#caffe_how_to">How to use the script</a></li>
<li class="level2"><a href="#caffe_result">What is the expected output from the script</a></li>
</ul>
</li>
<li class="level1"><a href="#tensorflow_data_extractor">Extract data from pre-trained tensorflow model</a><ul><li class="level2"><a href="#tensorflow_how_to">How to use the script</a></li>
<li class="level2"><a href="#tensorflow_result">What is the expected output from the script</a></li>
</ul>
</li>
</ul>
</div>
<div class="textblock"><h1><a class="anchor" id="caffe_data_extractor"></a>
Extract data from pre-trained caffe model</h1>
<p>One can find caffe <a href="https://github.com/BVLC/caffe/wiki/Model-Zoo">pre-trained models</a> on caffe's official github repository.</p>
<p>The <a class="el" href="caffe__data__extractor_8py.xhtml">caffe_data_extractor.py</a> provided in the <a class="el" href="dir_2bdd315ee2dbfd8cefe64730e37333fd.xhtml">scripts</a> folder is an example script that shows how to extract the parameter values from a trained model.</p>
<dl class="section note"><dt>Note</dt><dd>complex networks might require altering the script to properly work.</dd></dl>
<h2><a class="anchor" id="caffe_how_to"></a>
How to use the script</h2>
<p>Install caffe following <a href="http://caffe.berkeleyvision.org/installation.html">caffe's document</a>. Make sure the pycaffe has been added into the PYTHONPATH.</p>
<p>Download the pre-trained caffe model.</p>
<p>Run the <a class="el" href="caffe__data__extractor_8py.xhtml">caffe_data_extractor.py</a> script by </p>
<pre class="fragment"> python caffe_data_extractor.py -m &lt;caffe model&gt; -n &lt;caffe netlist&gt;
</pre><p>For example, to extract the data from pre-trained caffe Alex model to binary file: </p>
<pre class="fragment"> python caffe_data_extractor.py -m /path/to/bvlc_alexnet.caffemodel -n /path/to/caffe/models/bvlc_alexnet/deploy.prototxt
</pre><p>The script has been tested under Python2.7.</p>
<h2><a class="anchor" id="caffe_result"></a>
What is the expected output from the script</h2>
<p>If the script runs successfully, it prints the names and shapes of each layer onto the standard output and generates *.npy files containing the weights and biases of each layer.</p>
<p>The <a class="el" href="namespacearm__compute_1_1utils.xhtml#af214346f90d640ac468dd90fa2a275cc">arm_compute::utils::load_trained_data</a> shows how one could load the weights and biases into tensor from the .npy file by the help of Accessor.</p>
<h1><a class="anchor" id="tensorflow_data_extractor"></a>
Extract data from pre-trained tensorflow model</h1>
<p>The script <a class="el" href="tensorflow__data__extractor_8py.xhtml">tensorflow_data_extractor.py</a> extracts trainable parameters (e.g. values of weights and biases) from a trained tensorflow model. A tensorflow model consists of the following two files:</p>
<p>{model_name}.data-{step}-{global_step}: A binary file containing values of each variable.</p>
<p>{model_name}.meta: A binary file containing a MetaGraph struct which defines the graph structure of the neural network.</p>
<dl class="section note"><dt>Note</dt><dd>Since Tensorflow version 0.11 the binary checkpoint file which contains the values for each parameter has the format of: {model_name}.data-{step}-of-{max_step} instead of: {model_name}.ckpt When dealing with binary files with version &gt;= 0.11, only pass {model_name} to -m option; when dealing with binary files with version &lt; 0.11, pass the whole file name {model_name}.ckpt to -m option.</dd>
<dd>
This script relies on the parameters to be extracted being in the 'trainable_variables' tensor collection. By default all variables are automatically added to this collection unless specified otherwise by the user. Thus should a user alter this default behavior and/or want to extract parameters from other collections, tf.GraphKeys.TRAINABLE_VARIABLES should be replaced accordingly.</dd></dl>
<h2><a class="anchor" id="tensorflow_how_to"></a>
How to use the script</h2>
<p>Install tensorflow and numpy.</p>
<p>Download the pre-trained tensorflow model.</p>
<p>Run <a class="el" href="tensorflow__data__extractor_8py.xhtml">tensorflow_data_extractor.py</a> with </p>
<pre class="fragment"> python tensorflow_data_extractor -m &lt;path_to_binary_checkpoint_file&gt; -n &lt;path_to_metagraph_file&gt;
</pre><p>For example, to extract the data from pre-trained tensorflow Alex model to binary files: </p>
<pre class="fragment"> python tensorflow_data_extractor -m /path/to/bvlc_alexnet -n /path/to/bvlc_alexnet.meta
</pre><p>Or for binary checkpoint files before Tensorflow 0.11: </p>
<pre class="fragment"> python tensorflow_data_extractor -m /path/to/bvlc_alexnet.ckpt -n /path/to/bvlc_alexnet.meta
</pre><dl class="section note"><dt>Note</dt><dd>with versions &gt;= Tensorflow 0.11 only model name is passed to the -m option</dd></dl>
<p>The script has been tested with Tensorflow 1.2, 1.3 on Python 2.7.6 and Python 3.4.3.</p>
<h2><a class="anchor" id="tensorflow_result"></a>
What is the expected output from the script</h2>
<p>If the script runs successfully, it prints the names and shapes of each parameter onto the standard output and generates .npy files containing the weights and biases of each layer.</p>
<p>The <a class="el" href="namespacearm__compute_1_1utils.xhtml#af214346f90d640ac468dd90fa2a275cc">arm_compute::utils::load_trained_data</a> shows how one could load the weights and biases into tensor from the .npy file by the help of Accessor. </p>
</div></div><!-- contents -->
</div><!-- doc-content -->
<!-- start footer part -->
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
<ul>
<li class="footer">Generated on Thu Sep 28 2017 14:37:54 for Compute Library by
<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.6 </li>
</ul>
</div>
</body>
</html>