Google Git
Sign in
apache / mxnet-site / d5c5a235eef72f64ccaf224b834a20ef3287469b / . / versions / 1.9.1 / api / python / docs / tutorials / packages / onnx / fine_tuning_gluon.html
blob: 50edeb2d06cfdc528ab7aa52b17351545d9d46a3 [file] [log] [blame]
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta charset="utf-8" />
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no">
<meta http-equiv="x-ua-compatible" content="ie=edge">
<style>
.dropdown {
position: relative;
display: inline-block;
}
.dropdown-content {
display: none;
position: absolute;
background-color: #f9f9f9;
min-width: 160px;
box-shadow: 0px 8px 16px 0px rgba(0,0,0,0.2);
padding: 12px 16px;
z-index: 1;
text-align: left;
}
.dropdown:hover .dropdown-content {
display: block;
}
.dropdown-option:hover {
color: #FF4500;
}
.dropdown-option-active {
color: #FF4500;
font-weight: lighter;
}
.dropdown-option {
color: #000000;
font-weight: lighter;
}
.dropdown-header {
color: #FFFFFF;
display: inline-flex;
}
.dropdown-caret {
width: 18px;
}
.dropdown-caret-path {
fill: #FFFFFF;
}
</style>
<title>Fine-tuning an ONNX model &#8212; Apache MXNet documentation</title>
<link rel="stylesheet" href="../../../_static/basic.css" type="text/css" />
<link rel="stylesheet" href="../../../_static/pygments.css" type="text/css" />
<link rel="stylesheet" type="text/css" href="../../../_static/mxnet.css" />
<link rel="stylesheet" href="../../../_static/material-design-lite-1.3.0/material.blue-deep_orange.min.css" type="text/css" />
<link rel="stylesheet" href="../../../_static/sphinx_materialdesign_theme.css" type="text/css" />
<link rel="stylesheet" href="../../../_static/fontawesome/all.css" type="text/css" />
<link rel="stylesheet" href="../../../_static/fonts.css" type="text/css" />
<link rel="stylesheet" href="../../../_static/feedback.css" type="text/css" />
<script id="documentation_options" data-url_root="../../../" src="../../../_static/documentation_options.js"></script>
<script src="../../../_static/jquery.js"></script>
<script src="../../../_static/underscore.js"></script>
<script src="../../../_static/doctools.js"></script>
<script src="../../../_static/language_data.js"></script>
<script src="../../../_static/matomo_analytics.js"></script>
<script src="../../../_static/autodoc.js"></script>
<script crossorigin="anonymous" integrity="sha256-Ae2Vz/4ePdIu6ZyI/5ZGsYnb+m0JlOmKPjt6XZ9JJkA=" src="https://cdnjs.cloudflare.com/ajax/libs/require.js/2.3.4/require.min.js"></script>
<script async="async" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/latest.js?config=TeX-AMS-MML_HTMLorMML"></script>
<script type="text/x-mathjax-config">MathJax.Hub.Config({"tex2jax": {"inlineMath": [["$", "$"], ["\\(", "\\)"]], "processEscapes": true, "ignoreClass": "document", "processClass": "math|output_area"}})</script>
<link rel="shortcut icon" href="../../../_static/mxnet-icon.png"/>
<link rel="index" title="Index" href="../../../genindex.html" />
<link rel="search" title="Search" href="../../../search.html" />
<link rel="next" title="Running inference on MXNet/Gluon from an ONNX model" href="inference_on_onnx_model.html" />
<link rel="prev" title="ONNX" href="index.html" />
</head>
<body><header class="site-header" role="banner">
<div class="wrapper">
<a class="site-title" rel="author" href="/versions/1.9.1/"><img
src="../../../_static/mxnet_logo.png" class="site-header-logo"></a>
<nav class="site-nav">
<input type="checkbox" id="nav-trigger" class="nav-trigger"/>
<label for="nav-trigger">
<span class="menu-icon">
<svg viewBox="0 0 18 15" width="18px" height="15px">
<path d="M18,1.484c0,0.82-0.665,1.484-1.484,1.484H1.484C0.665,2.969,0,2.304,0,1.484l0,0C0,0.665,0.665,0,1.484,0 h15.032C17.335,0,18,0.665,18,1.484L18,1.484z M18,7.516C18,8.335,17.335,9,16.516,9H1.484C0.665,9,0,8.335,0,7.516l0,0 c0-0.82,0.665-1.484,1.484-1.484h15.032C17.335,6.031,18,6.696,18,7.516L18,7.516z M18,13.516C18,14.335,17.335,15,16.516,15H1.484 C0.665,15,0,14.335,0,13.516l0,0c0-0.82,0.665-1.483,1.484-1.483h15.032C17.335,12.031,18,12.695,18,13.516L18,13.516z"/>
</svg>
</span>
</label>
<div class="trigger">
<a class="page-link" href="/versions/1.9.1/get_started">Get Started</a>
<a class="page-link" href="/versions/1.9.1/features">Features</a>
<a class="page-link" href="/versions/1.9.1/ecosystem">Ecosystem</a>
<a class="page-link page-current" href="/versions/1.9.1/api">Docs & Tutorials</a>
<a class="page-link" href="/versions/1.9.1/trusted_by">Trusted By</a>
<a class="page-link" href="https://github.com/apache/mxnet">GitHub</a>
<div class="dropdown" style="min-width:100px">
<span class="dropdown-header">Apache
<svg class="dropdown-caret" viewBox="0 0 32 32" class="icon icon-caret-bottom" aria-hidden="true"><path class="dropdown-caret-path" d="M24 11.305l-7.997 11.39L8 11.305z"></path></svg>
</span>
<div class="dropdown-content" style="min-width:250px">
<a href="https://www.apache.org/foundation/">Apache Software Foundation</a>
<a href="https://incubator.apache.org/">Apache Incubator</a>
<a href="https://www.apache.org/licenses/">License</a>
<a href="/versions/1.9.1/api/faq/security.html">Security</a>
<a href="https://privacy.apache.org/policies/privacy-policy-public.html">Privacy</a>
<a href="https://www.apache.org/events/current-event">Events</a>
<a href="https://www.apache.org/foundation/sponsorship.html">Sponsorship</a>
<a href="https://www.apache.org/foundation/thanks.html">Thanks</a>
</div>
</div>
<div class="dropdown">
<span class="dropdown-header">1.9.1
<svg class="dropdown-caret" viewBox="0 0 32 32" class="icon icon-caret-bottom" aria-hidden="true"><path class="dropdown-caret-path" d="M24 11.305l-7.997 11.39L8 11.305z"></path></svg>
</span>
<div class="dropdown-content">
<a class="dropdown-option" href="/">master</a><br>
<a class="dropdown-option-active" href="/versions/1.9.1/">1.9.1</a><br>
<a class="dropdown-option" href="/versions/1.8.0/">1.8.0</a><br>
<a class="dropdown-option" href="/versions/1.7.0/">1.7.0</a><br>
<a class="dropdown-option" href="/versions/1.6.0/">1.6.0</a><br>
<a class="dropdown-option" href="/versions/1.5.0/">1.5.0</a><br>
<a class="dropdown-option" href="/versions/1.4.1/">1.4.1</a><br>
<a class="dropdown-option" href="/versions/1.3.1/">1.3.1</a><br>
<a class="dropdown-option" href="/versions/1.2.1/">1.2.1</a><br>
<a class="dropdown-option" href="/versions/1.1.0/">1.1.0</a><br>
<a class="dropdown-option" href="/versions/1.0.0/">1.0.0</a><br>
<a class="dropdown-option" href="/versions/0.12.1/">0.12.1</a><br>
<a class="dropdown-option" href="/versions/0.11.0/">0.11.0</a>
</div>
</div>
</div>
</nav>
</div>
</header>
<div class="mdl-layout mdl-js-layout mdl-layout--fixed-header mdl-layout--fixed-drawer"><header class="mdl-layout__header mdl-layout__header--waterfall ">
<div class="mdl-layout__header-row">
<nav class="mdl-navigation breadcrumb">
<a class="mdl-navigation__link" href="../../index.html">Python Tutorials</a><i class="material-icons">navigate_next</i>
<a class="mdl-navigation__link" href="../index.html">Packages</a><i class="material-icons">navigate_next</i>
<a class="mdl-navigation__link" href="index.html">ONNX</a><i class="material-icons">navigate_next</i>
<a class="mdl-navigation__link is-active">Fine-tuning an ONNX model</a>
</nav>
<div class="mdl-layout-spacer"></div>
<nav class="mdl-navigation">
<form class="form-inline pull-sm-right" action="../../../search.html" method="get">
<div class="mdl-textfield mdl-js-textfield mdl-textfield--expandable mdl-textfield--floating-label mdl-textfield--align-right">
<label id="quick-search-icon" class="mdl-button mdl-js-button mdl-button--icon" for="waterfall-exp">
<i class="material-icons">search</i>
</label>
<div class="mdl-textfield__expandable-holder">
<input class="mdl-textfield__input" type="text" name="q" id="waterfall-exp" placeholder="Search" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</div>
</div>
<div class="mdl-tooltip" data-mdl-for="quick-search-icon">
Quick search
</div>
</form>
<a id="button-show-source"
class="mdl-button mdl-js-button mdl-button--icon"
href="../../../_sources/tutorials/packages/onnx/fine_tuning_gluon.ipynb" rel="nofollow">
<i class="material-icons">code</i>
</a>
<div class="mdl-tooltip" data-mdl-for="button-show-source">
Show Source
</div>
</nav>
</div>
<div class="mdl-layout__header-row header-links">
<div class="mdl-layout-spacer"></div>
<nav class="mdl-navigation">
</nav>
</div>
</header><header class="mdl-layout__drawer">
<div class="globaltoc">
<span class="mdl-layout-title toc">Table Of Contents</span>
<nav class="mdl-navigation">
<ul class="current">
<li class="toctree-l1 current"><a class="reference internal" href="../../index.html">Python Tutorials</a><ul class="current">
<li class="toctree-l2"><a class="reference internal" href="../../getting-started/index.html">Getting Started</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/crash-course/index.html">Crash Course</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/1-ndarray.html">Manipulate data with <code class="docutils literal notranslate"><span class="pre">ndarray</span></code></a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/2-nn.html">Create a neural network</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/3-autograd.html">Automatic differentiation with <code class="docutils literal notranslate"><span class="pre">autograd</span></code></a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/4-train.html">Train the neural network</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/5-predict.html">Predict with a pre-trained model</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/6-use_gpus.html">Use GPUs</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/to-mxnet/index.html">Moving to MXNet from Other Frameworks</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/to-mxnet/pytorch.html">PyTorch vs Apache MXNet</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/gluon_from_experiment_to_deployment.html">Gluon: from experiment to deployment</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/logistic_regression_explained.html">Logistic regression explained</a></li>
<li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/image/mnist.html">MNIST</a></li>
</ul>
</li>
<li class="toctree-l2 current"><a class="reference internal" href="../index.html">Packages</a><ul class="current">
<li class="toctree-l3"><a class="reference internal" href="../autograd/index.html">Automatic Differentiation</a></li>
<li class="toctree-l3"><a class="reference internal" href="../gluon/index.html">Gluon</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../gluon/blocks/index.html">Blocks</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/custom-layer.html">Custom Layers</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/custom_layer_beginners.html">Customer Layers (Beginners)</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/hybridize.html">Hybridize</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/init.html">Initialization</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/naming.html">Parameter and Block Naming</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/nn.html">Layers and Blocks</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/parameters.html">Parameter Management</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/save_load_params.html">Saving and Loading Gluon Models</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/activations/activations.html">Activation Blocks</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/data/index.html">Data Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html">Image Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html#Spatial-Augmentation">Spatial Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html#Color-Augmentation">Color Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html#Composed-Augmentations">Composed Augmentations</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html">Gluon <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s and <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html#Using-own-data-with-included-Datasets">Using own data with included <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html#Using-own-data-with-custom-Datasets">Using own data with custom <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html#Appendix:-Upgrading-from-Module-DataIter-to-Gluon-DataLoader">Appendix: Upgrading from Module <code class="docutils literal notranslate"><span class="pre">DataIter</span></code> to Gluon <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/image/index.html">Image Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/image-augmentation.html">Image Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/info_gan.html">Image similarity search with InfoGAN</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/mnist.html">Handwritten Digit Recognition</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/pretrained_models.html">Using pre-trained models in MXNet</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/loss/index.html">Losses</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/loss/custom-loss.html">Custom Loss Blocks</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/loss/kl_divergence.html">Kullback-Leibler (KL) Divergence</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/loss/loss.html">Loss functions</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/text/index.html">Text Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/text/gnmt.html">Google Neural Machine Translation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/text/transformer.html">Machine Translation with Transformer</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/training/index.html">Training</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/fit_api_tutorial.html">MXNet Gluon Fit API</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/trainer.html">Trainer</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/learning_rates/index.html">Learning Rates</a><ul>
<li class="toctree-l6"><a class="reference internal" href="../gluon/training/learning_rates/learning_rate_finder.html">Learning Rate Finder</a></li>
<li class="toctree-l6"><a class="reference internal" href="../gluon/training/learning_rates/learning_rate_schedules.html">Learning Rate Schedules</a></li>
<li class="toctree-l6"><a class="reference internal" href="../gluon/training/learning_rates/learning_rate_schedules_advanced.html">Advanced Learning Rate Schedules</a></li>
</ul>
</li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/normalization/index.html">Normalization Blocks</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../kvstore/index.html">KVStore</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../kvstore/kvstore.html">Distributed Key-Value Store</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../ndarray/index.html">NDArray</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/01-ndarray-intro.html">An Intro: Manipulate Data the MXNet Way with NDArray</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/02-ndarray-operations.html">NDArray Operations</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/03-ndarray-contexts.html">NDArray Contexts</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/gotchas_numpy_in_mxnet.html">Gotchas using NumPy in Apache MXNet</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/sparse/index.html">Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/csr.html">CSRNDArray - NDArray in Compressed Sparse Row Storage Format</a></li>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/row_sparse.html">RowSparseNDArray - NDArray for Sparse Gradient Updates</a></li>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/train.html">Train a Linear Regression Model with Sparse Symbols</a></li>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/train_gluon.html">Sparse NDArrays with Gluon</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l3 current"><a class="reference internal" href="index.html">ONNX</a><ul class="current">
<li class="toctree-l4 current"><a class="current reference internal" href="#">Fine-tuning an ONNX model</a></li>
<li class="toctree-l4"><a class="reference internal" href="inference_on_onnx_model.html">Running inference on MXNet/Gluon from an ONNX model</a></li>
<li class="toctree-l4"><a class="reference internal" href="super_resolution.html">Importing an ONNX model into MXNet</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/deploy/export/onnx.html">Export ONNX Models</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../optimizer/index.html">Optimizers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../viz/index.html">Visualization</a><ul>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/visualize_graph">Visualize networks</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../performance/index.html">Performance</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../performance/compression/index.html">Compression</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../performance/compression/int8.html">Deploy with int-8</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/float16">Float16</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/gradient_compression">Gradient Compression</a></li>
<li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/int8_inference.html">GluonCV with Quantized Models</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../performance/backend/index.html">Accelerated Backend Tools</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/mkldnn/index.html">Intel MKL-DNN</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/mkldnn/mkldnn_quantization.html">Quantize with MKL-DNN backend</a></li>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/mkldnn/mkldnn_quantization.html#Improving-accuracy-with-Intel®-Neural-Compressor">Improving accuracy with Intel® Neural Compressor</a></li>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/mkldnn/mkldnn_readme.html">Install MXNet with MKL-DNN</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/tensorrt/index.html">TensorRT</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/tensorrt/tensorrt.html">Optimizing Deep Learning Computation Graphs with TensorRT</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/tvm.html">Use TVM</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/profiler.html">Profiling MXNet Models</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/amp.html">Using AMP: Automatic Mixed Precision</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../deploy/index.html">Deployment</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../deploy/export/index.html">Export</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/export/onnx.html">Exporting to ONNX format</a></li>
<li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/export_network.html">Export Gluon CV Models</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/blocks/save_load_params.html">Save / Load Parameters</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../deploy/inference/index.html">Inference</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/cpp.html">Deploy into C++</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/image_classification_jetson.html">Image Classication using pretrained ResNet-50 model on Jetson module</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/scala.html">Deploy into a Java or Scala Environment</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/wine_detector.html">Real-time Object Detection with MXNet On The Raspberry Pi</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../deploy/run-on-aws/index.html">Run on AWS</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/run-on-aws/use_ec2.html">Run on an EC2 Instance</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/run-on-aws/use_sagemaker.html">Run on Amazon SageMaker</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/run-on-aws/cloud.html">MXNet on the Cloud</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../extend/index.html">Extend</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../extend/custom_layer.html">Custom Layers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../extend/customop.html">Custom Numpy Operators</a></li>
<li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/new_op">New Operator Creation</a></li>
<li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/add_op_in_backend">New Operator in MXNet Backend</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../../api/index.html">Python API</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../../api/ndarray/index.html">mxnet.ndarray</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/ndarray.html">ndarray</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/contrib/index.html">ndarray.contrib</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/image/index.html">ndarray.image</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/linalg/index.html">ndarray.linalg</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/op/index.html">ndarray.op</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/random/index.html">ndarray.random</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/register/index.html">ndarray.register</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/sparse/index.html">ndarray.sparse</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/utils/index.html">ndarray.utils</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/gluon/index.html">mxnet.gluon</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/block.html">gluon.Block</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/hybrid_block.html">gluon.HybridBlock</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/symbol_block.html">gluon.SymbolBlock</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/constant.html">gluon.Constant</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter.html">gluon.Parameter</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter_dict.html">gluon.ParameterDict</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/trainer.html">gluon.Trainer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/contrib/index.html">gluon.contrib</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/data/index.html">gluon.data</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../../api/gluon/data/vision/index.html">data.vision</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/datasets/index.html">vision.datasets</a></li>
<li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/transforms/index.html">vision.transforms</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/loss/index.html">gluon.loss</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/model_zoo/index.html">gluon.model_zoo.vision</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/nn/index.html">gluon.nn</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/rnn/index.html">gluon.rnn</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/utils/index.html">gluon.utils</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/autograd/index.html">mxnet.autograd</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/initializer/index.html">mxnet.initializer</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/optimizer/index.html">mxnet.optimizer</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/lr_scheduler/index.html">mxnet.lr_scheduler</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/metric/index.html">mxnet.metric</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/kvstore/index.html">mxnet.kvstore</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/symbol/index.html">mxnet.symbol</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/symbol.html">symbol</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/contrib/index.html">symbol.contrib</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/image/index.html">symbol.image</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/linalg/index.html">symbol.linalg</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/op/index.html">symbol.op</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/random/index.html">symbol.random</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/register/index.html">symbol.register</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/sparse/index.html">symbol.sparse</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/module/index.html">mxnet.module</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/contrib/index.html">mxnet.contrib</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/autograd/index.html">contrib.autograd</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/io/index.html">contrib.io</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/ndarray/index.html">contrib.ndarray</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/onnx/index.html">contrib.onnx</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/quantization/index.html">contrib.quantization</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/symbol/index.html">contrib.symbol</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorboard/index.html">contrib.tensorboard</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorrt/index.html">contrib.tensorrt</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/text/index.html">contrib.text</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/mxnet/index.html">mxnet</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/attribute/index.html">mxnet.attribute</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/base/index.html">mxnet.base</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/callback/index.html">mxnet.callback</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/context/index.html">mxnet.context</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/engine/index.html">mxnet.engine</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor/index.html">mxnet.executor</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor_manager/index.html">mxnet.executor_manager</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/image/index.html">mxnet.image</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/io/index.html">mxnet.io</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/kvstore_server/index.html">mxnet.kvstore_server</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/libinfo/index.html">mxnet.libinfo</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/log/index.html">mxnet.log</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/model/index.html">mxnet.model</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/monitor/index.html">mxnet.monitor</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/name/index.html">mxnet.name</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/notebook/index.html">mxnet.notebook</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/operator/index.html">mxnet.operator</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/profiler/index.html">mxnet.profiler</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/random/index.html">mxnet.random</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/recordio/index.html">mxnet.recordio</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/registry/index.html">mxnet.registry</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/rtc/index.html">mxnet.rtc</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/runtime/index.html">mxnet.runtime</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/test_utils/index.html">mxnet.test_utils</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/torch/index.html">mxnet.torch</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/util/index.html">mxnet.util</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/visualization/index.html">mxnet.visualization</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</nav>
</div>
</header>
<main class="mdl-layout__content" tabIndex="0">
<script type="text/javascript" src="../../../_static/sphinx_materialdesign_theme.js "></script>
<script type="text/javascript" src="../../../_static/feedback.js"></script>
<header class="mdl-layout__drawer">
<div class="globaltoc">
<span class="mdl-layout-title toc">Table Of Contents</span>
<nav class="mdl-navigation">
<ul class="current">
<li class="toctree-l1 current"><a class="reference internal" href="../../index.html">Python Tutorials</a><ul class="current">
<li class="toctree-l2"><a class="reference internal" href="../../getting-started/index.html">Getting Started</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/crash-course/index.html">Crash Course</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/1-ndarray.html">Manipulate data with <code class="docutils literal notranslate"><span class="pre">ndarray</span></code></a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/2-nn.html">Create a neural network</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/3-autograd.html">Automatic differentiation with <code class="docutils literal notranslate"><span class="pre">autograd</span></code></a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/4-train.html">Train the neural network</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/5-predict.html">Predict with a pre-trained model</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/crash-course/6-use_gpus.html">Use GPUs</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/to-mxnet/index.html">Moving to MXNet from Other Frameworks</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../getting-started/to-mxnet/pytorch.html">PyTorch vs Apache MXNet</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/gluon_from_experiment_to_deployment.html">Gluon: from experiment to deployment</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../getting-started/logistic_regression_explained.html">Logistic regression explained</a></li>
<li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/image/mnist.html">MNIST</a></li>
</ul>
</li>
<li class="toctree-l2 current"><a class="reference internal" href="../index.html">Packages</a><ul class="current">
<li class="toctree-l3"><a class="reference internal" href="../autograd/index.html">Automatic Differentiation</a></li>
<li class="toctree-l3"><a class="reference internal" href="../gluon/index.html">Gluon</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../gluon/blocks/index.html">Blocks</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/custom-layer.html">Custom Layers</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/custom_layer_beginners.html">Customer Layers (Beginners)</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/hybridize.html">Hybridize</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/init.html">Initialization</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/naming.html">Parameter and Block Naming</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/nn.html">Layers and Blocks</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/parameters.html">Parameter Management</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/save_load_params.html">Saving and Loading Gluon Models</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/blocks/activations/activations.html">Activation Blocks</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/data/index.html">Data Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html">Image Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html#Spatial-Augmentation">Spatial Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html#Color-Augmentation">Color Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/data_augmentation.html#Composed-Augmentations">Composed Augmentations</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html">Gluon <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s and <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html#Using-own-data-with-included-Datasets">Using own data with included <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html#Using-own-data-with-custom-Datasets">Using own data with custom <code class="docutils literal notranslate"><span class="pre">Dataset</span></code>s</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/data/datasets.html#Appendix:-Upgrading-from-Module-DataIter-to-Gluon-DataLoader">Appendix: Upgrading from Module <code class="docutils literal notranslate"><span class="pre">DataIter</span></code> to Gluon <code class="docutils literal notranslate"><span class="pre">DataLoader</span></code></a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/image/index.html">Image Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/image-augmentation.html">Image Augmentation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/info_gan.html">Image similarity search with InfoGAN</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/mnist.html">Handwritten Digit Recognition</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/image/pretrained_models.html">Using pre-trained models in MXNet</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/loss/index.html">Losses</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/loss/custom-loss.html">Custom Loss Blocks</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/loss/kl_divergence.html">Kullback-Leibler (KL) Divergence</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/loss/loss.html">Loss functions</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/text/index.html">Text Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/text/gnmt.html">Google Neural Machine Translation</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/text/transformer.html">Machine Translation with Transformer</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../gluon/training/index.html">Training</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/fit_api_tutorial.html">MXNet Gluon Fit API</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/trainer.html">Trainer</a></li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/learning_rates/index.html">Learning Rates</a><ul>
<li class="toctree-l6"><a class="reference internal" href="../gluon/training/learning_rates/learning_rate_finder.html">Learning Rate Finder</a></li>
<li class="toctree-l6"><a class="reference internal" href="../gluon/training/learning_rates/learning_rate_schedules.html">Learning Rate Schedules</a></li>
<li class="toctree-l6"><a class="reference internal" href="../gluon/training/learning_rates/learning_rate_schedules_advanced.html">Advanced Learning Rate Schedules</a></li>
</ul>
</li>
<li class="toctree-l5"><a class="reference internal" href="../gluon/training/normalization/index.html">Normalization Blocks</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../kvstore/index.html">KVStore</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../kvstore/kvstore.html">Distributed Key-Value Store</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../ndarray/index.html">NDArray</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/01-ndarray-intro.html">An Intro: Manipulate Data the MXNet Way with NDArray</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/02-ndarray-operations.html">NDArray Operations</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/03-ndarray-contexts.html">NDArray Contexts</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/gotchas_numpy_in_mxnet.html">Gotchas using NumPy in Apache MXNet</a></li>
<li class="toctree-l4"><a class="reference internal" href="../ndarray/sparse/index.html">Tutorials</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/csr.html">CSRNDArray - NDArray in Compressed Sparse Row Storage Format</a></li>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/row_sparse.html">RowSparseNDArray - NDArray for Sparse Gradient Updates</a></li>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/train.html">Train a Linear Regression Model with Sparse Symbols</a></li>
<li class="toctree-l5"><a class="reference internal" href="../ndarray/sparse/train_gluon.html">Sparse NDArrays with Gluon</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l3 current"><a class="reference internal" href="index.html">ONNX</a><ul class="current">
<li class="toctree-l4 current"><a class="current reference internal" href="#">Fine-tuning an ONNX model</a></li>
<li class="toctree-l4"><a class="reference internal" href="inference_on_onnx_model.html">Running inference on MXNet/Gluon from an ONNX model</a></li>
<li class="toctree-l4"><a class="reference internal" href="super_resolution.html">Importing an ONNX model into MXNet</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/deploy/export/onnx.html">Export ONNX Models</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../optimizer/index.html">Optimizers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../viz/index.html">Visualization</a><ul>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/visualize_graph">Visualize networks</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../performance/index.html">Performance</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../performance/compression/index.html">Compression</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../performance/compression/int8.html">Deploy with int-8</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/float16">Float16</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/faq/gradient_compression">Gradient Compression</a></li>
<li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/int8_inference.html">GluonCV with Quantized Models</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../performance/backend/index.html">Accelerated Backend Tools</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/mkldnn/index.html">Intel MKL-DNN</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/mkldnn/mkldnn_quantization.html">Quantize with MKL-DNN backend</a></li>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/mkldnn/mkldnn_quantization.html#Improving-accuracy-with-Intel®-Neural-Compressor">Improving accuracy with Intel® Neural Compressor</a></li>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/mkldnn/mkldnn_readme.html">Install MXNet with MKL-DNN</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/tensorrt/index.html">TensorRT</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../../performance/backend/tensorrt/tensorrt.html">Optimizing Deep Learning Computation Graphs with TensorRT</a></li>
</ul>
</li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/tvm.html">Use TVM</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/profiler.html">Profiling MXNet Models</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../performance/backend/amp.html">Using AMP: Automatic Mixed Precision</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../deploy/index.html">Deployment</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../deploy/export/index.html">Export</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/export/onnx.html">Exporting to ONNX format</a></li>
<li class="toctree-l4"><a class="reference external" href="https://gluon-cv.mxnet.io/build/examples_deployment/export_network.html">Export Gluon CV Models</a></li>
<li class="toctree-l4"><a class="reference external" href="https://mxnet.apache.org/api/python/docs/tutorials/packages/gluon/blocks/save_load_params.html">Save / Load Parameters</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../deploy/inference/index.html">Inference</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/cpp.html">Deploy into C++</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/image_classification_jetson.html">Image Classication using pretrained ResNet-50 model on Jetson module</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/scala.html">Deploy into a Java or Scala Environment</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/inference/wine_detector.html">Real-time Object Detection with MXNet On The Raspberry Pi</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../deploy/run-on-aws/index.html">Run on AWS</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/run-on-aws/use_ec2.html">Run on an EC2 Instance</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/run-on-aws/use_sagemaker.html">Run on Amazon SageMaker</a></li>
<li class="toctree-l4"><a class="reference internal" href="../../deploy/run-on-aws/cloud.html">MXNet on the Cloud</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../extend/index.html">Extend</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../extend/custom_layer.html">Custom Layers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../extend/customop.html">Custom Numpy Operators</a></li>
<li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/new_op">New Operator Creation</a></li>
<li class="toctree-l3"><a class="reference external" href="https://mxnet.apache.org/api/faq/add_op_in_backend">New Operator in MXNet Backend</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../../api/index.html">Python API</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../../api/ndarray/index.html">mxnet.ndarray</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/ndarray.html">ndarray</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/contrib/index.html">ndarray.contrib</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/image/index.html">ndarray.image</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/linalg/index.html">ndarray.linalg</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/op/index.html">ndarray.op</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/random/index.html">ndarray.random</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/register/index.html">ndarray.register</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/sparse/index.html">ndarray.sparse</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/ndarray/utils/index.html">ndarray.utils</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/gluon/index.html">mxnet.gluon</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/block.html">gluon.Block</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/hybrid_block.html">gluon.HybridBlock</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/symbol_block.html">gluon.SymbolBlock</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/constant.html">gluon.Constant</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter.html">gluon.Parameter</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/parameter_dict.html">gluon.ParameterDict</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/trainer.html">gluon.Trainer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/contrib/index.html">gluon.contrib</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/data/index.html">gluon.data</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../../../api/gluon/data/vision/index.html">data.vision</a><ul>
<li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/datasets/index.html">vision.datasets</a></li>
<li class="toctree-l5"><a class="reference internal" href="../../../api/gluon/data/vision/transforms/index.html">vision.transforms</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/loss/index.html">gluon.loss</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/model_zoo/index.html">gluon.model_zoo.vision</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/nn/index.html">gluon.nn</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/rnn/index.html">gluon.rnn</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/gluon/utils/index.html">gluon.utils</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/autograd/index.html">mxnet.autograd</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/initializer/index.html">mxnet.initializer</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/optimizer/index.html">mxnet.optimizer</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/lr_scheduler/index.html">mxnet.lr_scheduler</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/metric/index.html">mxnet.metric</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/kvstore/index.html">mxnet.kvstore</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/symbol/index.html">mxnet.symbol</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/symbol.html">symbol</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/contrib/index.html">symbol.contrib</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/image/index.html">symbol.image</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/linalg/index.html">symbol.linalg</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/op/index.html">symbol.op</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/random/index.html">symbol.random</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/register/index.html">symbol.register</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/symbol/sparse/index.html">symbol.sparse</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/module/index.html">mxnet.module</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/contrib/index.html">mxnet.contrib</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/autograd/index.html">contrib.autograd</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/io/index.html">contrib.io</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/ndarray/index.html">contrib.ndarray</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/onnx/index.html">contrib.onnx</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/quantization/index.html">contrib.quantization</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/symbol/index.html">contrib.symbol</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorboard/index.html">contrib.tensorboard</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/tensorrt/index.html">contrib.tensorrt</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/contrib/text/index.html">contrib.text</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../../api/mxnet/index.html">mxnet</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/attribute/index.html">mxnet.attribute</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/base/index.html">mxnet.base</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/callback/index.html">mxnet.callback</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/context/index.html">mxnet.context</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/engine/index.html">mxnet.engine</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor/index.html">mxnet.executor</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/executor_manager/index.html">mxnet.executor_manager</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/image/index.html">mxnet.image</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/io/index.html">mxnet.io</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/kvstore_server/index.html">mxnet.kvstore_server</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/libinfo/index.html">mxnet.libinfo</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/log/index.html">mxnet.log</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/model/index.html">mxnet.model</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/monitor/index.html">mxnet.monitor</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/name/index.html">mxnet.name</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/notebook/index.html">mxnet.notebook</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/operator/index.html">mxnet.operator</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/profiler/index.html">mxnet.profiler</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/random/index.html">mxnet.random</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/recordio/index.html">mxnet.recordio</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/registry/index.html">mxnet.registry</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/rtc/index.html">mxnet.rtc</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/runtime/index.html">mxnet.runtime</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/test_utils/index.html">mxnet.test_utils</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/torch/index.html">mxnet.torch</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/util/index.html">mxnet.util</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../../api/mxnet/visualization/index.html">mxnet.visualization</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</nav>
</div>
</header>
<div class="document">
<div class="page-content" role="main">
<!--- Licensed to the Apache Software Foundation (ASF) under one --><!--- or more contributor license agreements. See the NOTICE file --><!--- distributed with this work for additional information --><!--- regarding copyright ownership. The ASF licenses this file --><!--- to you under the Apache License, Version 2.0 (the --><!--- "License"); you may not use this file except in compliance --><!--- with the License. You may obtain a copy of the License at --><!--- http://www.apache.org/licenses/LICENSE-2.0 --><!--- Unless required by applicable law or agreed to in writing, --><!--- software distributed under the License is distributed on an --><!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY --><!--- KIND, either express or implied. See the License for the --><!--- specific language governing permissions and limitations --><!--- under the License. --><div class="section" id="Fine-tuning-an-ONNX-model">
<h1>Fine-tuning an ONNX model<a class="headerlink" href="#Fine-tuning-an-ONNX-model" title="Permalink to this headline"></a></h1>
<p>Fine-tuning is a common practice in Transfer Learning. One can take advantage of the pre-trained weights of a network, and use them as an initializer for their own task. Indeed, quite often it is difficult to gather a dataset large enough that it would allow training from scratch deep and complex networks such as ResNet152 or VGG16. For example in an image classification task, using a network trained on a large dataset like ImageNet gives a good base from which the weights can be slightly
updated, or fine-tuned, to predict accurately the new classes. We will see in this tutorial that this can be achieved even with a relatively small number of new training examples.</p>
<p><a class="reference external" href="https://github.com/onnx/onnx">Open Neural Network Exchange (ONNX)</a> provides an open source format for AI models. It defines an extensible computation graph model, as well as definitions of built-in operators and standard data types.</p>
<p>In this tutorial we will:</p>
<ul class="simple">
<li><p>learn how to pick a specific layer from a pre-trained .onnx model file</p></li>
<li><p>learn how to load this model in Gluon and fine-tune it on a different dataset</p></li>
</ul>
<div class="section" id="Pre-requisite">
<h2>Pre-requisite<a class="headerlink" href="#Pre-requisite" title="Permalink to this headline"></a></h2>
<p>To run the tutorial you will need to have installed the following python modules: - <a class="reference external" href="/get_started">MXNet &gt; 1.1.0</a> - <a class="reference external" href="https://github.com/onnx/onnx">onnx</a> - matplotlib</p>
<p>We recommend that you have first followed this tutorial: - <a class="reference external" href="/api/python/docs/tutorials/packages/onnx/inference_on_onnx_model.html">Inference using an ONNX model on MXNet Gluon</a></p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">json</span>
<span class="kn">import</span> <span class="nn">logging</span>
<span class="kn">import</span> <span class="nn">multiprocessing</span>
<span class="kn">import</span> <span class="nn">os</span>
<span class="kn">import</span> <span class="nn">tarfile</span>
<span class="n">logging</span><span class="o">.</span><span class="n">basicConfig</span><span class="p">(</span><span class="n">level</span><span class="o">=</span><span class="n">logging</span><span class="o">.</span><span class="n">INFO</span><span class="p">)</span>
<span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="nn">plt</span>
<span class="kn">import</span> <span class="nn">mxnet</span> <span class="k">as</span> <span class="nn">mx</span>
<span class="kn">from</span> <span class="nn">mxnet</span> <span class="kn">import</span> <span class="n">gluon</span><span class="p">,</span> <span class="n">nd</span><span class="p">,</span> <span class="n">autograd</span>
<span class="kn">from</span> <span class="nn">mxnet.gluon.data.vision.datasets</span> <span class="kn">import</span> <span class="n">ImageFolderDataset</span>
<span class="kn">from</span> <span class="nn">mxnet.gluon.data</span> <span class="kn">import</span> <span class="n">DataLoader</span>
<span class="kn">import</span> <span class="nn">mxnet.contrib.onnx</span> <span class="k">as</span> <span class="nn">onnx_mxnet</span>
<span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="o">%</span><span class="n">matplotlib</span> <span class="n">inline</span>
</pre></div>
</div>
<div class="section" id="Downloading-supporting-files">
<h3>Downloading supporting files<a class="headerlink" href="#Downloading-supporting-files" title="Permalink to this headline"></a></h3>
<p>These are images and a vizualisation script:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">image_folder</span> <span class="o">=</span> <span class="s2">&quot;images&quot;</span>
<span class="n">utils_file</span> <span class="o">=</span> <span class="s2">&quot;utils.py&quot;</span> <span class="c1"># contain utils function to plot nice visualization</span>
<span class="n">images</span> <span class="o">=</span> <span class="p">[</span><span class="s1">&#39;wrench.jpg&#39;</span><span class="p">,</span> <span class="s1">&#39;dolphin.jpg&#39;</span><span class="p">,</span> <span class="s1">&#39;lotus.jpg&#39;</span><span class="p">]</span>
<span class="n">base_url</span> <span class="o">=</span> <span class="s2">&quot;https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/doc/tutorials/onnx/</span><span class="si">{}</span><span class="s2">?raw=true&quot;</span>
<span class="k">for</span> <span class="n">image</span> <span class="ow">in</span> <span class="n">images</span><span class="p">:</span>
<span class="n">mx</span><span class="o">.</span><span class="n">test_utils</span><span class="o">.</span><span class="n">download</span><span class="p">(</span><span class="n">base_url</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="s2">&quot;</span><span class="si">{}</span><span class="s2">/</span><span class="si">{}</span><span class="s2">&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">image_folder</span><span class="p">,</span> <span class="n">image</span><span class="p">)),</span> <span class="n">fname</span><span class="o">=</span><span class="n">image</span><span class="p">,</span><span class="n">dirname</span><span class="o">=</span><span class="n">image_folder</span><span class="p">)</span>
<span class="n">mx</span><span class="o">.</span><span class="n">test_utils</span><span class="o">.</span><span class="n">download</span><span class="p">(</span><span class="n">base_url</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">utils_file</span><span class="p">),</span> <span class="n">fname</span><span class="o">=</span><span class="n">utils_file</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">utils</span> <span class="kn">import</span> <span class="o">*</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="Downloading-a-model-from-the-ONNX-model-zoo">
<h2>Downloading a model from the ONNX model zoo<a class="headerlink" href="#Downloading-a-model-from-the-ONNX-model-zoo" title="Permalink to this headline"></a></h2>
<p>We download a pre-trained model, in our case the <a class="reference external" href="https://arxiv.org/abs/1409.4842">GoogleNet</a> model, trained on <a class="reference external" href="http://www.image-net.org/">ImageNet</a> from the <a class="reference external" href="https://github.com/onnx/models">ONNX model zoo</a>. The model comes packaged in an archive <code class="docutils literal notranslate"><span class="pre">tar.gz</span></code> file containing an <code class="docutils literal notranslate"><span class="pre">model.onnx</span></code> model file.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">base_url</span> <span class="o">=</span> <span class="s2">&quot;https://s3.amazonaws.com/download.onnx/models/opset_3/&quot;</span>
<span class="n">current_model</span> <span class="o">=</span> <span class="s2">&quot;bvlc_googlenet&quot;</span>
<span class="n">model_folder</span> <span class="o">=</span> <span class="s2">&quot;model&quot;</span>
<span class="n">archive_file</span> <span class="o">=</span> <span class="s2">&quot;</span><span class="si">{}</span><span class="s2">.tar.gz&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">current_model</span><span class="p">)</span>
<span class="n">archive_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">model_folder</span><span class="p">,</span> <span class="n">archive_file</span><span class="p">)</span>
<span class="n">url</span> <span class="o">=</span> <span class="s2">&quot;</span><span class="si">{}{}</span><span class="s2">&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">base_url</span><span class="p">,</span> <span class="n">archive_file</span><span class="p">)</span>
<span class="n">onnx_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">model_folder</span><span class="p">,</span> <span class="n">current_model</span><span class="p">,</span> <span class="s1">&#39;model.onnx&#39;</span><span class="p">)</span>
<span class="c1"># Download the zipped model</span>
<span class="n">mx</span><span class="o">.</span><span class="n">test_utils</span><span class="o">.</span><span class="n">download</span><span class="p">(</span><span class="n">url</span><span class="p">,</span> <span class="n">dirname</span> <span class="o">=</span> <span class="n">model_folder</span><span class="p">)</span>
<span class="c1"># Extract the model</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">isdir</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">model_folder</span><span class="p">,</span> <span class="n">current_model</span><span class="p">)):</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;Extracting </span><span class="si">{}</span><span class="s1"> in </span><span class="si">{}</span><span class="s1">...&#39;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">archive_path</span><span class="p">,</span> <span class="n">model_folder</span><span class="p">))</span>
<span class="n">tar</span> <span class="o">=</span> <span class="n">tarfile</span><span class="o">.</span><span class="n">open</span><span class="p">(</span><span class="n">archive_path</span><span class="p">,</span> <span class="s2">&quot;r:gz&quot;</span><span class="p">)</span>
<span class="n">tar</span><span class="o">.</span><span class="n">extractall</span><span class="p">(</span><span class="n">model_folder</span><span class="p">)</span>
<span class="n">tar</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;Model extracted.&#39;</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="Downloading-the-Caltech101-dataset">
<h2>Downloading the Caltech101 dataset<a class="headerlink" href="#Downloading-the-Caltech101-dataset" title="Permalink to this headline"></a></h2>
<p>The <a class="reference external" href="http://www.vision.caltech.edu/Image_Datasets/Caltech101/">Caltech101 dataset</a> is made of pictures of objects belonging to 101 categories. About 40 to 800 images per category. Most categories have about 50 images.</p>
<p><em>L. Fei-Fei, R. Fergus and P. Perona. Learning generative visual models from few training examples: an incremental Bayesian approach tested on 101 object categories. IEEE. CVPR 2004, Workshop on Generative-Model Based Vision. 2004</em></p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">data_folder</span> <span class="o">=</span> <span class="s2">&quot;data&quot;</span>
<span class="n">dataset_name</span> <span class="o">=</span> <span class="s2">&quot;101_ObjectCategories&quot;</span>
<span class="n">archive_file</span> <span class="o">=</span> <span class="s2">&quot;</span><span class="si">{}</span><span class="s2">.tar.gz&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">dataset_name</span><span class="p">)</span>
<span class="n">archive_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">data_folder</span><span class="p">,</span> <span class="n">archive_file</span><span class="p">)</span>
<span class="n">data_url</span> <span class="o">=</span> <span class="s2">&quot;https://s3.us-east-2.amazonaws.com/mxnet-public/&quot;</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">isfile</span><span class="p">(</span><span class="n">archive_path</span><span class="p">):</span>
<span class="n">mx</span><span class="o">.</span><span class="n">test_utils</span><span class="o">.</span><span class="n">download</span><span class="p">(</span><span class="s2">&quot;</span><span class="si">{}{}</span><span class="s2">&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">data_url</span><span class="p">,</span> <span class="n">archive_file</span><span class="p">),</span> <span class="n">dirname</span> <span class="o">=</span> <span class="n">data_folder</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;Extracting </span><span class="si">{}</span><span class="s1"> in </span><span class="si">{}</span><span class="s1">...&#39;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">archive_file</span><span class="p">,</span> <span class="n">data_folder</span><span class="p">))</span>
<span class="n">tar</span> <span class="o">=</span> <span class="n">tarfile</span><span class="o">.</span><span class="n">open</span><span class="p">(</span><span class="n">archive_path</span><span class="p">,</span> <span class="s2">&quot;r:gz&quot;</span><span class="p">)</span>
<span class="n">tar</span><span class="o">.</span><span class="n">extractall</span><span class="p">(</span><span class="n">data_folder</span><span class="p">)</span>
<span class="n">tar</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;Data extracted.&#39;</span><span class="p">)</span>
</pre></div>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">training_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">data_folder</span><span class="p">,</span> <span class="n">dataset_name</span><span class="p">)</span>
<span class="n">testing_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">data_folder</span><span class="p">,</span> <span class="s2">&quot;</span><span class="si">{}</span><span class="s2">_test&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">dataset_name</span><span class="p">))</span>
</pre></div>
</div>
<div class="section" id="Load-the-data-using-an-ImageFolderDataset-and-a-DataLoader">
<h3>Load the data using an ImageFolderDataset and a DataLoader<a class="headerlink" href="#Load-the-data-using-an-ImageFolderDataset-and-a-DataLoader" title="Permalink to this headline"></a></h3>
<p>We need to transform the images to a format accepted by the network</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">EDGE</span> <span class="o">=</span> <span class="mi">224</span>
<span class="n">SIZE</span> <span class="o">=</span> <span class="p">(</span><span class="n">EDGE</span><span class="p">,</span> <span class="n">EDGE</span><span class="p">)</span>
<span class="n">BATCH_SIZE</span> <span class="o">=</span> <span class="mi">32</span>
<span class="n">NUM_WORKERS</span> <span class="o">=</span> <span class="mi">6</span>
</pre></div>
</div>
<p>We transform the dataset images using the following operations: - resize the shorter edge to 224, the longer edge will be greater or equal to 224 - center and crop an area of size (224,224) - transpose the channels to be (3,224,224)</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">transform</span><span class="p">(</span><span class="n">image</span><span class="p">,</span> <span class="n">label</span><span class="p">):</span>
<span class="n">resized</span> <span class="o">=</span> <span class="n">mx</span><span class="o">.</span><span class="n">image</span><span class="o">.</span><span class="n">resize_short</span><span class="p">(</span><span class="n">image</span><span class="p">,</span> <span class="n">EDGE</span><span class="p">)</span>
<span class="n">cropped</span><span class="p">,</span> <span class="n">crop_info</span> <span class="o">=</span> <span class="n">mx</span><span class="o">.</span><span class="n">image</span><span class="o">.</span><span class="n">center_crop</span><span class="p">(</span><span class="n">resized</span><span class="p">,</span> <span class="n">SIZE</span><span class="p">)</span>
<span class="n">transposed</span> <span class="o">=</span> <span class="n">nd</span><span class="o">.</span><span class="n">transpose</span><span class="p">(</span><span class="n">cropped</span><span class="p">,</span> <span class="p">(</span><span class="mi">2</span><span class="p">,</span><span class="mi">0</span><span class="p">,</span><span class="mi">1</span><span class="p">))</span>
<span class="k">return</span> <span class="n">transposed</span><span class="p">,</span> <span class="n">label</span>
</pre></div>
</div>
<p>The train and test dataset are created automatically by passing the root of each folder. The labels are built using the sub-folders names as label.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">train_root</span>
<span class="n">__label1</span>
<span class="n">____image1</span>
<span class="n">____image2</span>
<span class="n">__label2</span>
<span class="n">____image3</span>
<span class="n">____image4</span>
</pre></div>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">dataset_train</span> <span class="o">=</span> <span class="n">ImageFolderDataset</span><span class="p">(</span><span class="n">root</span><span class="o">=</span><span class="n">training_path</span><span class="p">)</span>
<span class="n">dataset_test</span> <span class="o">=</span> <span class="n">ImageFolderDataset</span><span class="p">(</span><span class="n">root</span><span class="o">=</span><span class="n">testing_path</span><span class="p">)</span>
</pre></div>
</div>
<p>We use several worker processes, which means the dataloading and pre-processing is going to be distributed across multiple processes. This will help preventing our GPU from starving and waiting for the data to be copied across</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">dataloader_train</span> <span class="o">=</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">dataset_train</span><span class="o">.</span><span class="n">transform</span><span class="p">(</span><span class="n">transform</span><span class="p">,</span> <span class="n">lazy</span><span class="o">=</span><span class="kc">False</span><span class="p">),</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">BATCH_SIZE</span><span class="p">,</span> <span class="n">last_batch</span><span class="o">=</span><span class="s1">&#39;rollover&#39;</span><span class="p">,</span>
<span class="n">shuffle</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">num_workers</span><span class="o">=</span><span class="n">NUM_WORKERS</span><span class="p">)</span>
<span class="n">dataloader_test</span> <span class="o">=</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">dataset_test</span><span class="o">.</span><span class="n">transform</span><span class="p">(</span><span class="n">transform</span><span class="p">,</span> <span class="n">lazy</span><span class="o">=</span><span class="kc">False</span><span class="p">),</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">BATCH_SIZE</span><span class="p">,</span> <span class="n">last_batch</span><span class="o">=</span><span class="s1">&#39;rollover&#39;</span><span class="p">,</span>
<span class="n">shuffle</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">num_workers</span><span class="o">=</span><span class="n">NUM_WORKERS</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">&quot;Train dataset: </span><span class="si">{}</span><span class="s2"> images, Test dataset: </span><span class="si">{}</span><span class="s2"> images&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">dataset_train</span><span class="p">),</span> <span class="nb">len</span><span class="p">(</span><span class="n">dataset_test</span><span class="p">)))</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">Train</span> <span class="pre">dataset:</span> <span class="pre">6996</span> <span class="pre">images,</span> <span class="pre">Test</span> <span class="pre">dataset:</span> <span class="pre">1681</span> <span class="pre">images</span></code></p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">categories</span> <span class="o">=</span> <span class="n">dataset_train</span><span class="o">.</span><span class="n">synsets</span>
<span class="n">NUM_CLASSES</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">categories</span><span class="p">)</span>
<span class="n">BATCH_SIZE</span> <span class="o">=</span> <span class="mi">32</span>
</pre></div>
</div>
<p>Let’s plot the 1000th image to test the dataset</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">N</span> <span class="o">=</span> <span class="mi">1000</span>
<span class="n">plt</span><span class="o">.</span><span class="n">imshow</span><span class="p">((</span><span class="n">transform</span><span class="p">(</span><span class="n">dataset_train</span><span class="p">[</span><span class="n">N</span><span class="p">][</span><span class="mi">0</span><span class="p">],</span> <span class="mi">0</span><span class="p">)[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">asnumpy</span><span class="p">()</span><span class="o">.</span><span class="n">transpose</span><span class="p">((</span><span class="mi">1</span><span class="p">,</span><span class="mi">2</span><span class="p">,</span><span class="mi">0</span><span class="p">))))</span>
<span class="n">plt</span><span class="o">.</span><span class="n">axis</span><span class="p">(</span><span class="s1">&#39;off&#39;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">categories</span><span class="p">[</span><span class="n">dataset_train</span><span class="p">[</span><span class="n">N</span><span class="p">][</span><span class="mi">1</span><span class="p">]])</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">Motorbikes</span></code></p>
<p><img alt="png" src="https://github.com/dmlc/web-data/blob/master/mxnet/doc/tutorials/onnx/caltech101_correct.png?raw=true" /></p>
</div>
</div>
<div class="section" id="Fine-Tuning-the-ONNX-model">
<h2>Fine-Tuning the ONNX model<a class="headerlink" href="#Fine-Tuning-the-ONNX-model" title="Permalink to this headline"></a></h2>
<div class="section" id="Getting-the-last-layer">
<h3>Getting the last layer<a class="headerlink" href="#Getting-the-last-layer" title="Permalink to this headline"></a></h3>
<p>Load the ONNX model</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">sym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span> <span class="o">=</span> <span class="n">onnx_mxnet</span><span class="o">.</span><span class="n">import_model</span><span class="p">(</span><span class="n">onnx_path</span><span class="p">)</span>
</pre></div>
</div>
<p>This function get the output of a given layer</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">get_layer_output</span><span class="p">(</span><span class="n">symbol</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span> <span class="n">layer_name</span><span class="p">):</span>
<span class="n">all_layers</span> <span class="o">=</span> <span class="n">symbol</span><span class="o">.</span><span class="n">get_internals</span><span class="p">()</span>
<span class="n">net</span> <span class="o">=</span> <span class="n">all_layers</span><span class="p">[</span><span class="n">layer_name</span><span class="o">+</span><span class="s1">&#39;_output&#39;</span><span class="p">]</span>
<span class="n">net</span> <span class="o">=</span> <span class="n">mx</span><span class="o">.</span><span class="n">symbol</span><span class="o">.</span><span class="n">Flatten</span><span class="p">(</span><span class="n">data</span><span class="o">=</span><span class="n">net</span><span class="p">)</span>
<span class="n">new_args</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">({</span><span class="n">k</span><span class="p">:</span><span class="n">arg_params</span><span class="p">[</span><span class="n">k</span><span class="p">]</span> <span class="k">for</span> <span class="n">k</span> <span class="ow">in</span> <span class="n">arg_params</span> <span class="k">if</span> <span class="n">k</span> <span class="ow">in</span> <span class="n">net</span><span class="o">.</span><span class="n">list_arguments</span><span class="p">()})</span>
<span class="n">new_aux</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">({</span><span class="n">k</span><span class="p">:</span><span class="n">aux_params</span><span class="p">[</span><span class="n">k</span><span class="p">]</span> <span class="k">for</span> <span class="n">k</span> <span class="ow">in</span> <span class="n">aux_params</span> <span class="k">if</span> <span class="n">k</span> <span class="ow">in</span> <span class="n">net</span><span class="o">.</span><span class="n">list_arguments</span><span class="p">()})</span>
<span class="k">return</span> <span class="p">(</span><span class="n">net</span><span class="p">,</span> <span class="n">new_args</span><span class="p">,</span> <span class="n">new_aux</span><span class="p">)</span>
</pre></div>
</div>
<p>Here we print the different layers of the network to make it easier to pick the right one</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">sym</span><span class="o">.</span><span class="n">get_internals</span><span class="p">()</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">&lt;Symbol</span> <span class="pre">group</span> <span class="pre">[data_0,</span> <span class="pre">pad0,</span> <span class="pre">conv1/7x7_s2_w_0,</span> <span class="pre">conv1/7x7_s2_b_0,</span> <span class="pre">convolution0,</span> <span class="pre">relu0,</span> <span class="pre">pad1,</span> <span class="pre">pooling0,</span> <span class="pre">lrn0,</span> <span class="pre">pad2,</span> <span class="pre">conv2/3x3_reduce_w_0,</span> <span class="pre">conv2/3x3_reduce_b_0,</span> <span class="pre">convolution1,</span> <span class="pre">relu1,</span> <span class="pre">pad3,</span> <span class="pre">conv2/3x3_w_0,</span> <span class="pre">conv2/3x3_b_0,</span> <span class="pre">convolution2,</span> <span class="pre">relu2,</span> <span class="pre">lrn1,</span> <span class="pre">pad4,</span> <span class="pre">pooling1,</span> <span class="pre">pad5,</span> <span class="pre">inception_3a/1x1_w_0,</span> <span class="pre">inception_3a/1x1_b_0,</span> <span class="pre">convolution3,</span> <span class="pre">relu3,</span> <span class="pre">pad6,</span> <span class="pre">.................................................................................inception_5b/pool_proj_b_0,</span> <span class="pre">convolution56,</span> <span class="pre">relu56,</span> <span class="pre">concat8,</span> <span class="pre">pad70,</span> <span class="pre">pooling13,</span> <span class="pre">dropout0,</span> <span class="pre">flatten0,</span> <span class="pre">loss3/classifier_w_0,</span> <span class="pre">linalg_gemm20,</span> <span class="pre">loss3/classifier_b_0,</span> <span class="pre">_mulscalar0,</span> <span class="pre">broadcast_add0,</span> <span class="pre">softmax0]&gt;</span></code></p>
<p>We get the network until the output of the <code class="docutils literal notranslate"><span class="pre">flatten0</span></code> layer</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">new_sym</span><span class="p">,</span> <span class="n">new_arg_params</span><span class="p">,</span> <span class="n">new_aux_params</span> <span class="o">=</span> <span class="n">get_layer_output</span><span class="p">(</span><span class="n">sym</span><span class="p">,</span> <span class="n">arg_params</span><span class="p">,</span> <span class="n">aux_params</span><span class="p">,</span> <span class="s1">&#39;flatten0&#39;</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="Fine-tuning-in-gluon">
<h3>Fine-tuning in gluon<a class="headerlink" href="#Fine-tuning-in-gluon" title="Permalink to this headline"></a></h3>
<p>We can now take advantage of the features and pattern detection knowledge that our network learnt training on ImageNet, and apply that to the new Caltech101 dataset.</p>
<p>We pick a context, fine-tuning on CPU will be <strong>WAY</strong> slower.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">ctx</span> <span class="o">=</span> <span class="n">mx</span><span class="o">.</span><span class="n">gpu</span><span class="p">()</span> <span class="k">if</span> <span class="n">mx</span><span class="o">.</span><span class="n">context</span><span class="o">.</span><span class="n">num_gpus</span><span class="p">()</span> <span class="o">&gt;</span> <span class="mi">0</span> <span class="k">else</span> <span class="n">mx</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span>
</pre></div>
</div>
<p>We create a symbol block that is going to hold all our pre-trained layers, and assign the weights of the different pre-trained layers to the newly created SymbolBlock</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">warnings</span>
<span class="k">with</span> <span class="n">warnings</span><span class="o">.</span><span class="n">catch_warnings</span><span class="p">():</span>
<span class="n">warnings</span><span class="o">.</span><span class="n">simplefilter</span><span class="p">(</span><span class="s2">&quot;ignore&quot;</span><span class="p">)</span>
<span class="n">pre_trained</span> <span class="o">=</span> <span class="n">gluon</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">SymbolBlock</span><span class="p">(</span><span class="n">outputs</span><span class="o">=</span><span class="n">new_sym</span><span class="p">,</span> <span class="n">inputs</span><span class="o">=</span><span class="n">mx</span><span class="o">.</span><span class="n">sym</span><span class="o">.</span><span class="n">var</span><span class="p">(</span><span class="s1">&#39;data_0&#39;</span><span class="p">))</span>
<span class="n">net_params</span> <span class="o">=</span> <span class="n">pre_trained</span><span class="o">.</span><span class="n">collect_params</span><span class="p">()</span>
<span class="k">for</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">new_arg_params</span><span class="p">:</span>
<span class="k">if</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">net_params</span><span class="p">:</span>
<span class="n">net_params</span><span class="p">[</span><span class="n">param</span><span class="p">]</span><span class="o">.</span><span class="n">_load_init</span><span class="p">(</span><span class="n">new_arg_params</span><span class="p">[</span><span class="n">param</span><span class="p">],</span> <span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">)</span>
<span class="k">for</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">new_aux_params</span><span class="p">:</span>
<span class="k">if</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">net_params</span><span class="p">:</span>
<span class="n">net_params</span><span class="p">[</span><span class="n">param</span><span class="p">]</span><span class="o">.</span><span class="n">_load_init</span><span class="p">(</span><span class="n">new_aux_params</span><span class="p">[</span><span class="n">param</span><span class="p">],</span> <span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">)</span>
</pre></div>
</div>
<p>We create the new dense layer with the right new number of classes (101) and initialize the weights</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">dense_layer</span> <span class="o">=</span> <span class="n">gluon</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Dense</span><span class="p">(</span><span class="n">NUM_CLASSES</span><span class="p">)</span>
<span class="n">dense_layer</span><span class="o">.</span><span class="n">initialize</span><span class="p">(</span><span class="n">mx</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">Xavier</span><span class="p">(</span><span class="n">magnitude</span><span class="o">=</span><span class="mf">2.24</span><span class="p">),</span> <span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">)</span>
</pre></div>
</div>
<p>We add the SymbolBlock and the new dense layer to a HybridSequential network</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">net</span> <span class="o">=</span> <span class="n">gluon</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">HybridSequential</span><span class="p">()</span>
<span class="k">with</span> <span class="n">net</span><span class="o">.</span><span class="n">name_scope</span><span class="p">():</span>
<span class="n">net</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">pre_trained</span><span class="p">)</span>
<span class="n">net</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">dense_layer</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="Loss">
<h3>Loss<a class="headerlink" href="#Loss" title="Permalink to this headline"></a></h3>
<p>Softmax cross entropy for multi-class classification</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">softmax_cross_entropy</span> <span class="o">=</span> <span class="n">gluon</span><span class="o">.</span><span class="n">loss</span><span class="o">.</span><span class="n">SoftmaxCrossEntropyLoss</span><span class="p">()</span>
</pre></div>
</div>
</div>
<div class="section" id="Trainer">
<h3>Trainer<a class="headerlink" href="#Trainer" title="Permalink to this headline"></a></h3>
<p>Initialize trainer with common training parameters</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">LEARNING_RATE</span> <span class="o">=</span> <span class="mf">0.0005</span>
<span class="n">WDECAY</span> <span class="o">=</span> <span class="mf">0.00001</span>
<span class="n">MOMENTUM</span> <span class="o">=</span> <span class="mf">0.9</span>
</pre></div>
</div>
<p>The trainer will retrain and fine-tune the entire network. If we use <code class="docutils literal notranslate"><span class="pre">dense_layer</span></code> instead of <code class="docutils literal notranslate"><span class="pre">net</span></code> in the cell below, the gradient updates would only be applied to the new last dense layer. Essentially we would be using the pre-trained network as a featurizer.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">trainer</span> <span class="o">=</span> <span class="n">gluon</span><span class="o">.</span><span class="n">Trainer</span><span class="p">(</span><span class="n">net</span><span class="o">.</span><span class="n">collect_params</span><span class="p">(),</span> <span class="s1">&#39;sgd&#39;</span><span class="p">,</span>
<span class="p">{</span><span class="s1">&#39;learning_rate&#39;</span><span class="p">:</span> <span class="n">LEARNING_RATE</span><span class="p">,</span>
<span class="s1">&#39;wd&#39;</span><span class="p">:</span><span class="n">WDECAY</span><span class="p">,</span>
<span class="s1">&#39;momentum&#39;</span><span class="p">:</span><span class="n">MOMENTUM</span><span class="p">})</span>
</pre></div>
</div>
</div>
<div class="section" id="Evaluation-loop">
<h3>Evaluation loop<a class="headerlink" href="#Evaluation-loop" title="Permalink to this headline"></a></h3>
<p>We measure the accuracy in a non-blocking way, using <code class="docutils literal notranslate"><span class="pre">nd.array</span></code> to take care of the parallelisation that MXNet and Gluon offers.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">evaluate_accuracy_gluon</span><span class="p">(</span><span class="n">data_iterator</span><span class="p">,</span> <span class="n">net</span><span class="p">):</span>
<span class="n">num_instance</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">sum_metric</span> <span class="o">=</span> <span class="n">nd</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span><span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">)</span>
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">label</span><span class="p">)</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">data_iterator</span><span class="p">):</span>
<span class="n">data</span> <span class="o">=</span> <span class="n">data</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span><span class="o">.</span><span class="n">as_in_context</span><span class="p">(</span><span class="n">ctx</span><span class="p">)</span>
<span class="n">label</span> <span class="o">=</span> <span class="n">label</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">)</span><span class="o">.</span><span class="n">as_in_context</span><span class="p">(</span><span class="n">ctx</span><span class="p">)</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">net</span><span class="p">(</span><span class="n">data</span><span class="p">)</span>
<span class="n">prediction</span> <span class="o">=</span> <span class="n">nd</span><span class="o">.</span><span class="n">argmax</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">axis</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">)</span>
<span class="n">num_instance</span> <span class="o">+=</span> <span class="nb">len</span><span class="p">(</span><span class="n">prediction</span><span class="p">)</span>
<span class="n">sum_metric</span> <span class="o">+=</span> <span class="p">(</span><span class="n">prediction</span><span class="o">==</span><span class="n">label</span><span class="p">)</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
<span class="n">accuracy</span> <span class="o">=</span> <span class="p">(</span><span class="n">sum_metric</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span><span class="o">/</span><span class="n">num_instance</span><span class="p">)</span>
<span class="k">return</span> <span class="n">accuracy</span><span class="o">.</span><span class="n">asscalar</span><span class="p">()</span>
</pre></div>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="o">%%</span><span class="n">time</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">&quot;Untrained network Test Accuracy: </span><span class="si">{0:.4f}</span><span class="s2">&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">evaluate_accuracy_gluon</span><span class="p">(</span><span class="n">dataloader_test</span><span class="p">,</span> <span class="n">net</span><span class="p">)))</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">Untrained</span> <span class="pre">network</span> <span class="pre">Test</span> <span class="pre">Accuracy:</span> <span class="pre">0.0192</span></code></p>
</div>
<div class="section" id="Training-loop">
<h3>Training loop<a class="headerlink" href="#Training-loop" title="Permalink to this headline"></a></h3>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">val_accuracy</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">for</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">5</span><span class="p">):</span>
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">label</span><span class="p">)</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">dataloader_train</span><span class="p">):</span>
<span class="n">data</span> <span class="o">=</span> <span class="n">data</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span><span class="o">.</span><span class="n">as_in_context</span><span class="p">(</span><span class="n">ctx</span><span class="p">)</span>
<span class="n">label</span> <span class="o">=</span> <span class="n">label</span><span class="o">.</span><span class="n">as_in_context</span><span class="p">(</span><span class="n">ctx</span><span class="p">)</span>
<span class="k">if</span> <span class="n">i</span><span class="o">%</span><span class="mi">20</span><span class="o">==</span><span class="mi">0</span> <span class="ow">and</span> <span class="n">i</span> <span class="o">&gt;</span><span class="mi">0</span><span class="p">:</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;Batch [</span><span class="si">{0}</span><span class="s1">] loss: </span><span class="si">{1:.4f}</span><span class="s1">&#39;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">,</span> <span class="n">loss</span><span class="o">.</span><span class="n">mean</span><span class="p">()</span><span class="o">.</span><span class="n">asscalar</span><span class="p">()))</span>
<span class="k">with</span> <span class="n">autograd</span><span class="o">.</span><span class="n">record</span><span class="p">():</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">net</span><span class="p">(</span><span class="n">data</span><span class="p">)</span>
<span class="n">loss</span> <span class="o">=</span> <span class="n">softmax_cross_entropy</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">label</span><span class="p">)</span>
<span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="n">trainer</span><span class="o">.</span><span class="n">step</span><span class="p">(</span><span class="n">data</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
<span class="n">nd</span><span class="o">.</span><span class="n">waitall</span><span class="p">()</span> <span class="c1"># wait at the end of the epoch</span>
<span class="n">new_val_accuracy</span> <span class="o">=</span> <span class="n">evaluate_accuracy_gluon</span><span class="p">(</span><span class="n">dataloader_test</span><span class="p">,</span> <span class="n">net</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">&quot;Epoch [</span><span class="si">{0}</span><span class="s2">] Test Accuracy </span><span class="si">{1:.4f}</span><span class="s2"> &quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">epoch</span><span class="p">,</span> <span class="n">new_val_accuracy</span><span class="p">))</span>
<span class="c1"># We perform early-stopping regularization, to prevent the model from overfitting</span>
<span class="k">if</span> <span class="n">val_accuracy</span> <span class="o">&gt;</span> <span class="n">new_val_accuracy</span><span class="p">:</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;Validation accuracy is decreasing, stopping training&#39;</span><span class="p">)</span>
<span class="k">break</span>
<span class="n">val_accuracy</span> <span class="o">=</span> <span class="n">new_val_accuracy</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">Epoch</span> <span class="pre">4,</span> <span class="pre">Test</span> <span class="pre">Accuracy</span> <span class="pre">0.8942307829856873</span></code></p>
</div>
</div>
<div class="section" id="Testing">
<h2>Testing<a class="headerlink" href="#Testing" title="Permalink to this headline"></a></h2>
<p>In the previous tutorial, we saw that the network trained on ImageNet couldn’t classify correctly <code class="docutils literal notranslate"><span class="pre">wrench</span></code>, <code class="docutils literal notranslate"><span class="pre">dolphin</span></code>, <code class="docutils literal notranslate"><span class="pre">lotus</span></code> because these are not categories of the ImageNet dataset.</p>
<p>Let’s see if our network fine-tuned on Caltech101 is up for the task:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Number of predictions to show</span>
<span class="n">TOP_P</span> <span class="o">=</span> <span class="mi">3</span>
</pre></div>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Convert img to format expected by the network</span>
<span class="k">def</span> <span class="nf">transform</span><span class="p">(</span><span class="n">img</span><span class="p">):</span>
<span class="k">return</span> <span class="n">nd</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">expand_dims</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">transpose</span><span class="p">(</span><span class="n">img</span><span class="p">,</span> <span class="p">(</span><span class="mi">2</span><span class="p">,</span><span class="mi">0</span><span class="p">,</span><span class="mi">1</span><span class="p">)),</span><span class="n">axis</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span><span class="o">.</span><span class="n">astype</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">),</span> <span class="n">ctx</span><span class="o">=</span><span class="n">ctx</span><span class="p">)</span>
</pre></div>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Load and transform the test images</span>
<span class="n">caltech101_images_test</span> <span class="o">=</span> <span class="p">[</span><span class="n">plt</span><span class="o">.</span><span class="n">imread</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">image_folder</span><span class="p">,</span> <span class="s2">&quot;</span><span class="si">{}</span><span class="s2">&quot;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">img</span><span class="p">)))</span> <span class="k">for</span> <span class="n">img</span> <span class="ow">in</span> <span class="n">images</span><span class="p">]</span>
<span class="n">caltech101_images_transformed</span> <span class="o">=</span> <span class="p">[</span><span class="n">transform</span><span class="p">(</span><span class="n">img</span><span class="p">)</span> <span class="k">for</span> <span class="n">img</span> <span class="ow">in</span> <span class="n">caltech101_images_test</span><span class="p">]</span>
</pre></div>
</div>
<p>Helper function to run batches of data</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">run_batch</span><span class="p">(</span><span class="n">net</span><span class="p">,</span> <span class="n">data</span><span class="p">):</span>
<span class="n">results</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">batch</span> <span class="ow">in</span> <span class="n">data</span><span class="p">:</span>
<span class="n">outputs</span> <span class="o">=</span> <span class="n">net</span><span class="p">(</span><span class="n">batch</span><span class="p">)</span>
<span class="n">results</span><span class="o">.</span><span class="n">extend</span><span class="p">([</span><span class="n">o</span> <span class="k">for</span> <span class="n">o</span> <span class="ow">in</span> <span class="n">outputs</span><span class="o">.</span><span class="n">asnumpy</span><span class="p">()])</span>
<span class="k">return</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">results</span><span class="p">)</span>
</pre></div>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">result</span> <span class="o">=</span> <span class="n">run_batch</span><span class="p">(</span><span class="n">net</span><span class="p">,</span> <span class="n">caltech101_images_transformed</span><span class="p">)</span>
</pre></div>
</div>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">plot_predictions</span><span class="p">(</span><span class="n">caltech101_images_test</span><span class="p">,</span> <span class="n">result</span><span class="p">,</span> <span class="n">categories</span><span class="p">,</span> <span class="n">TOP_P</span><span class="p">)</span>
</pre></div>
</div>
<p><img alt="png" src="https://github.com/dmlc/web-data/blob/master/mxnet/doc/tutorials/onnx/caltech101_correct.png?raw=true" /></p>
<p><strong>Great!</strong> The network classified these images correctly after being fine-tuned on a dataset that contains images of <code class="docutils literal notranslate"><span class="pre">wrench</span></code>, <code class="docutils literal notranslate"><span class="pre">dolphin</span></code> and <code class="docutils literal notranslate"><span class="pre">lotus</span></code></p>
</div>
</div>
<hr class="feedback-hr-top" />
<div class="feedback-container">
<div class="feedback-question">Did this page help you?</div>
<div class="feedback-answer-container">
<div class="feedback-answer yes-link" data-response="yes">Yes</div>
<div class="feedback-answer no-link" data-response="no">No</div>
</div>
<div class="feedback-thank-you">Thanks for your feedback!</div>
</div>
<hr class="feedback-hr-bottom" />
</div>
<div class="side-doc-outline">
<div class="side-doc-outline--content">
<div class="localtoc">
<p class="caption">
<span class="caption-text">Table Of Contents</span>
</p>
<ul>
<li><a class="reference internal" href="#">Fine-tuning an ONNX model</a><ul>
<li><a class="reference internal" href="#Pre-requisite">Pre-requisite</a><ul>
<li><a class="reference internal" href="#Downloading-supporting-files">Downloading supporting files</a></li>
</ul>
</li>
<li><a class="reference internal" href="#Downloading-a-model-from-the-ONNX-model-zoo">Downloading a model from the ONNX model zoo</a></li>
<li><a class="reference internal" href="#Downloading-the-Caltech101-dataset">Downloading the Caltech101 dataset</a><ul>
<li><a class="reference internal" href="#Load-the-data-using-an-ImageFolderDataset-and-a-DataLoader">Load the data using an ImageFolderDataset and a DataLoader</a></li>
</ul>
</li>
<li><a class="reference internal" href="#Fine-Tuning-the-ONNX-model">Fine-Tuning the ONNX model</a><ul>
<li><a class="reference internal" href="#Getting-the-last-layer">Getting the last layer</a></li>
<li><a class="reference internal" href="#Fine-tuning-in-gluon">Fine-tuning in gluon</a></li>
<li><a class="reference internal" href="#Loss">Loss</a></li>
<li><a class="reference internal" href="#Trainer">Trainer</a></li>
<li><a class="reference internal" href="#Evaluation-loop">Evaluation loop</a></li>
<li><a class="reference internal" href="#Training-loop">Training loop</a></li>
</ul>
</li>
<li><a class="reference internal" href="#Testing">Testing</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
<div class="clearer"></div>
</div><div class="pagenation">
<a id="button-prev" href="index.html" class="mdl-button mdl-js-button mdl-js-ripple-effect mdl-button--colored" role="botton" accesskey="P">
<i class="pagenation-arrow-L fas fa-arrow-left fa-lg"></i>
<div class="pagenation-text">
<span class="pagenation-direction">Previous</span>
<div>ONNX</div>
</div>
</a>
<a id="button-next" href="inference_on_onnx_model.html" class="mdl-button mdl-js-button mdl-js-ripple-effect mdl-button--colored" role="botton" accesskey="N">
<i class="pagenation-arrow-R fas fa-arrow-right fa-lg"></i>
<div class="pagenation-text">
<span class="pagenation-direction">Next</span>
<div>Running inference on MXNet/Gluon from an ONNX model</div>
</div>
</a>
</div>
<footer class="site-footer h-card">
<div class="wrapper">
<div class="row">
<div class="col-4">
<h4 class="footer-category-title">Resources</h4>
<ul class="contact-list">
<li><a class="u-email" href="mailto:dev@mxnet.apache.org">Dev list</a></li>
<li><a class="u-email" href="mailto:user@mxnet.apache.org">User mailing list</a></li>
<li><a href="https://cwiki.apache.org/confluence/display/MXNET/Apache+MXNet+Home">Developer Wiki</a></li>
<li><a href="https://issues.apache.org/jira/projects/MXNET/issues">Jira Tracker</a></li>
<li><a href="https://github.com/apache/mxnet/labels/Roadmap">Github Roadmap</a></li>
<li><a href="https://medium.com/apache-mxnet">Blog</a></li>
<li><a href="https://discuss.mxnet.io">Forum</a></li>
<li><a href="/community/contribute">Contribute</a></li>
</ul>
</div>
<div class="col-4"><ul class="social-media-list"><li><a href="https://github.com/apache/mxnet"><svg class="svg-icon"><use xlink:href="../../../_static/minima-social-icons.svg#github"></use></svg> <span class="username">apache/mxnet</span></a></li><li><a href="https://www.twitter.com/apachemxnet"><svg class="svg-icon"><use xlink:href="../../../_static/minima-social-icons.svg#twitter"></use></svg> <span class="username">apachemxnet</span></a></li><li><a href="https://youtube.com/apachemxnet"><svg class="svg-icon"><use xlink:href="../../../_static/minima-social-icons.svg#youtube"></use></svg> <span class="username">apachemxnet</span></a></li></ul>
</div>
<div class="col-4 footer-text">
<p>A flexible and efficient library for deep learning.</p>
</div>
</div>
</div>
</footer>
<footer class="site-footer2">
<div class="wrapper">
<div class="row">
<div class="col-3">
<img src="../../../_static/apache_incubator_logo.png" class="footer-logo col-2">
</div>
<div class="footer-bottom-warning col-9">
<p>Apache MXNet is an effort undergoing incubation at <a href="http://www.apache.org/">The Apache Software Foundation</a> (ASF), <span style="font-weight:bold">sponsored by the <i>Apache Incubator</i></span>. Incubation is required
of all newly accepted projects until a further review indicates that the infrastructure,
communications, and decision making process have stabilized in a manner consistent with other
successful ASF projects. While incubation status is not necessarily a reflection of the completeness
or stability of the code, it does indicate that the project has yet to be fully endorsed by the ASF.
</p><p>"Copyright © 2017-2018, The Apache Software Foundation Apache MXNet, MXNet, Apache, the Apache
feather, and the Apache MXNet project logo are either registered trademarks or trademarks of the
Apache Software Foundation."</p>
</div>
</div>
</div>
</footer>
</body>
</html>