mirror of
https://github.com/labmlai/annotated_deep_learning_paper_implementations.git
synced 2025-08-14 01:13:00 +08:00
1006 lines
61 KiB
HTML
1006 lines
61 KiB
HTML
<!DOCTYPE html>
|
|
<html>
|
|
<head>
|
|
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
|
|
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
|
<meta name="description" content="A PyTorch implementation/tutorial of Deep Residual Learning for Image Recognition (ResNet)."/>
|
|
|
|
<meta name="twitter:card" content="summary"/>
|
|
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
|
<meta name="twitter:title" content="Deep Residual Learning for Image Recognition (ResNet)"/>
|
|
<meta name="twitter:description" content="A PyTorch implementation/tutorial of Deep Residual Learning for Image Recognition (ResNet)."/>
|
|
<meta name="twitter:site" content="@labmlai"/>
|
|
<meta name="twitter:creator" content="@labmlai"/>
|
|
|
|
<meta property="og:url" content="https://nn.labml.ai/resnet/index.html"/>
|
|
<meta property="og:title" content="Deep Residual Learning for Image Recognition (ResNet)"/>
|
|
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
|
<meta property="og:site_name" content="LabML Neural Networks"/>
|
|
<meta property="og:type" content="object"/>
|
|
<meta property="og:title" content="Deep Residual Learning for Image Recognition (ResNet)"/>
|
|
<meta property="og:description" content="A PyTorch implementation/tutorial of Deep Residual Learning for Image Recognition (ResNet)."/>
|
|
|
|
<title>Deep Residual Learning for Image Recognition (ResNet)</title>
|
|
<link rel="shortcut icon" href="/icon.png"/>
|
|
<link rel="stylesheet" href="../pylit.css">
|
|
<link rel="canonical" href="https://nn.labml.ai/resnet/index.html"/>
|
|
<!-- Global site tag (gtag.js) - Google Analytics -->
|
|
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
|
|
<script>
|
|
window.dataLayer = window.dataLayer || [];
|
|
|
|
function gtag() {
|
|
dataLayer.push(arguments);
|
|
}
|
|
|
|
gtag('js', new Date());
|
|
|
|
gtag('config', 'G-4V3HC8HBLH');
|
|
</script>
|
|
</head>
|
|
<body>
|
|
<div id='container'>
|
|
<div id="background"></div>
|
|
<div class='section'>
|
|
<div class='docs'>
|
|
<p>
|
|
<a class="parent" href="/">home</a>
|
|
<a class="parent" href="index.html">resnet</a>
|
|
</p>
|
|
<p>
|
|
|
|
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/resnet/__init__.py">
|
|
<img alt="Github"
|
|
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
|
|
style="max-width:100%;"/></a>
|
|
<a href="https://twitter.com/labmlai"
|
|
rel="nofollow">
|
|
<img alt="Twitter"
|
|
src="https://img.shields.io/twitter/follow/labmlai?style=social"
|
|
style="max-width:100%;"/></a>
|
|
</p>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-0'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-0'>#</a>
|
|
</div>
|
|
<h1>Deep Residual Learning for Image Recognition (ResNet)</h1>
|
|
<p>This is a <a href="https://pytorch.org">PyTorch</a> implementation of the paper
|
|
<a href="https://papers.labml.ai/paper/1512.03385">Deep Residual Learning for Image Recognition</a>.</p>
|
|
<p>ResNets train layers as residual functions to overcome the
|
|
<em>degradation problem</em>.
|
|
The degradation problem is the accuracy of deep neural networks degrading when
|
|
the number of layers becomes very high.
|
|
The accuracy increases as the number of layers increase, then saturates,
|
|
and then starts to degrade.</p>
|
|
<p>The paper argues that deeper models should perform at least as well as shallower
|
|
models because the extra layers can just learn to perform an identity mapping.</p>
|
|
<h2>Residual Learning</h2>
|
|
<p>If $\mathcal{H}(x)$ is the mapping that needs to be learned by a few layers,
|
|
they train the residual function</p>
|
|
<p>
|
|
<script type="math/tex; mode=display">\mathcal{F}(x) = \mathcal{H}(x) - x</script>
|
|
</p>
|
|
<p>instead. And the original function becomes $\mathcal{F}(x) + x$.</p>
|
|
<p>In this case, learning identity mapping for $\mathcal{H}(x)$ is
|
|
equivalent to learning $\mathcal{F}(x)$ to be $0$, which is easier to
|
|
learn.</p>
|
|
<p>In the parameterized form this can be written as,</p>
|
|
<p>
|
|
<script type="math/tex; mode=display">\mathcal{F}(x, \{W_i\}) + x</script>
|
|
</p>
|
|
<p>and when the feature map sizes of $\mathcal{F}(x, {W_i})$ and $x$ are different
|
|
the paper suggests doing a linear projection, with learned weights $W_s$.</p>
|
|
<p>
|
|
<script type="math/tex; mode=display">\mathcal{F}(x, \{W_i\}) + W_s x</script>
|
|
</p>
|
|
<p>Paper experimented with zero padding instead of linear projections and found linear projections
|
|
to work better. Also when the feature map sizes match they found identity mapping
|
|
to be better than linear projections.</p>
|
|
<p>$\mathcal{F}$ should have more than one layer, otherwise the sum $\mathcal{F}(x, {W_i}) + W_s x$
|
|
also won’t have non-linearities and will be like a linear layer.</p>
|
|
<p>Here is <a href="experiment.html">the training code</a> for training a ResNet on CIFAR-10.</p>
|
|
<p><a href="https://app.labml.ai/run/fc5ad600e4af11ebbafd23b8665193c1"><img alt="View Run" src="https://img.shields.io/badge/labml-experiment-brightgreen" /></a></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">57</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span><span class="p">,</span> <span class="n">Optional</span>
|
|
<span class="lineno">58</span>
|
|
<span class="lineno">59</span><span class="kn">import</span> <span class="nn">torch</span>
|
|
<span class="lineno">60</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
|
<span class="lineno">61</span>
|
|
<span class="lineno">62</span><span class="kn">from</span> <span class="nn">labml_helpers.module</span> <span class="kn">import</span> <span class="n">Module</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-1'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-1'>#</a>
|
|
</div>
|
|
<h2>Linear projections for shortcut connection</h2>
|
|
<p>This does the $W_s x$ projection described above.</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">65</span><span class="k">class</span> <span class="nc">ShortcutProjection</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-2'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-2'>#</a>
|
|
</div>
|
|
<ul>
|
|
<li><code>in_channels</code> is the number of channels in $x$</li>
|
|
<li><code>out_channels</code> is the number of channels in $\mathcal{F}(x, {W_i})$</li>
|
|
<li><code>stride</code> is the stride length in the convolution operation for $F$.
|
|
We do the same stride on the shortcut connection, to match the feature-map size.</li>
|
|
</ul>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">72</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">in_channels</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">stride</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-3'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-3'>#</a>
|
|
</div>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">79</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-4'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-4'>#</a>
|
|
</div>
|
|
<p>Convolution layer for linear projection $W_s x$</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">82</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="n">in_channels</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">,</span> <span class="n">kernel_size</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="n">stride</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-5'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-5'>#</a>
|
|
</div>
|
|
<p>Paper suggests adding batch normalization after each convolution operation</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">84</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="n">out_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-6'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-6'>#</a>
|
|
</div>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">86</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-7'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-7'>#</a>
|
|
</div>
|
|
<p>Convolution and batch normalization</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">88</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="p">(</span><span class="n">x</span><span class="p">))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-8'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-8'>#</a>
|
|
</div>
|
|
<p><a id="residual_block"></a></p>
|
|
<h2>Residual Block</h2>
|
|
<p>This implements the residual block described in the paper.
|
|
It has two $3 \times 3$ convolution layers.</p>
|
|
<p><img alt="Residual Block" src="residual_block.svg" /></p>
|
|
<p>The first convolution layer maps from <code>in_channels</code> to <code>out_channels</code>,
|
|
where the <code>out_channels</code> is higher than <code>in_channels</code> when we reduce the
|
|
feature map size with a stride length greater than $1$.</p>
|
|
<p>The second convolution layer maps from <code>out_channels</code> to <code>out_channels</code> and
|
|
always has a stride length of 1.</p>
|
|
<p>Both convolution layers are followed by batch normalization.</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">91</span><span class="k">class</span> <span class="nc">ResidualBlock</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-9'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-9'>#</a>
|
|
</div>
|
|
<ul>
|
|
<li><code>in_channels</code> is the number of channels in $x$</li>
|
|
<li><code>out_channels</code> is the number of output channels</li>
|
|
<li><code>stride</code> is the stride length in the convolution operation.</li>
|
|
</ul>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">111</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">in_channels</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">stride</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-10'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-10'>#</a>
|
|
</div>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">117</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-11'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-11'>#</a>
|
|
</div>
|
|
<p>First $3 \times 3$ convolution layer, this maps to <code>out_channels</code></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">120</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="n">in_channels</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">,</span> <span class="n">kernel_size</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="n">stride</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-12'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-12'>#</a>
|
|
</div>
|
|
<p>Batch normalization after the first convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">122</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="n">out_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-13'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-13'>#</a>
|
|
</div>
|
|
<p>First activation function (ReLU)</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">124</span> <span class="bp">self</span><span class="o">.</span><span class="n">act1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-14'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-14'>#</a>
|
|
</div>
|
|
<p>Second $3 \times 3$ convolution layer</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">127</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="n">out_channels</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">,</span> <span class="n">kernel_size</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-15'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-15'>#</a>
|
|
</div>
|
|
<p>Batch normalization after the second convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">129</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="n">out_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-16'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-16'>#</a>
|
|
</div>
|
|
<p>Shortcut connection should be a projection if the stride length is not $1$
|
|
of if the number of channels change</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">133</span> <span class="k">if</span> <span class="n">stride</span> <span class="o">!=</span> <span class="mi">1</span> <span class="ow">or</span> <span class="n">in_channels</span> <span class="o">!=</span> <span class="n">out_channels</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-17'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-17'>#</a>
|
|
</div>
|
|
<p>Projection $W_s x$</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">135</span> <span class="bp">self</span><span class="o">.</span><span class="n">shortcut</span> <span class="o">=</span> <span class="n">ShortcutProjection</span><span class="p">(</span><span class="n">in_channels</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">,</span> <span class="n">stride</span><span class="p">)</span>
|
|
<span class="lineno">136</span> <span class="k">else</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-18'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-18'>#</a>
|
|
</div>
|
|
<p>Identity $x$</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">138</span> <span class="bp">self</span><span class="o">.</span><span class="n">shortcut</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Identity</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-19'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-19'>#</a>
|
|
</div>
|
|
<p>Second activation function (ReLU) (after adding the shortcut)</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">141</span> <span class="bp">self</span><span class="o">.</span><span class="n">act2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-20'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-20'>#</a>
|
|
</div>
|
|
<ul>
|
|
<li><code>x</code> is the input of shape <code>[batch_size, in_channels, height, width]</code></li>
|
|
</ul>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">143</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-21'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-21'>#</a>
|
|
</div>
|
|
<p>Get the shortcut connection</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">148</span> <span class="n">shortcut</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">shortcut</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-22'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-22'>#</a>
|
|
</div>
|
|
<p>First convolution and activation</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">150</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">act1</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">bn1</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv1</span><span class="p">(</span><span class="n">x</span><span class="p">)))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-23'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-23'>#</a>
|
|
</div>
|
|
<p>Second convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">152</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn2</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv2</span><span class="p">(</span><span class="n">x</span><span class="p">))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-24'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-24'>#</a>
|
|
</div>
|
|
<p>Activation function after adding the shortcut</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">154</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">act2</span><span class="p">(</span><span class="n">x</span> <span class="o">+</span> <span class="n">shortcut</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-25'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-25'>#</a>
|
|
</div>
|
|
<p><a id="bottleneck_residual_block"></a></p>
|
|
<h2>Bottleneck Residual Block</h2>
|
|
<p>This implements the bottleneck block described in the paper.
|
|
It has $1 \times 1$, $3 \times 3$, and $1 \times 1$ convolution layers.</p>
|
|
<p><img alt="Bottlenext Block" src="bottleneck_block.svg" /></p>
|
|
<p>The first convolution layer maps from <code>in_channels</code> to <code>bottleneck_channels</code> with a $1x1$
|
|
convolution,
|
|
where the <code>bottleneck_channels</code> is lower than <code>in_channels</code>.</p>
|
|
<p>The second $3x3$ convolution layer maps from <code>bottleneck_channels</code> to <code>bottleneck_channels</code>.
|
|
This can have a stride length greater than $1$ when we want to compress the
|
|
feature map size.</p>
|
|
<p>The third, final $1x1$ convolution layer maps to <code>out_channels</code>.
|
|
<code>out_channels</code> is higher than <code>in_channels</code> if the stride length is greater than $1$;
|
|
otherwise, $out_channels$ is equal to <code>in_channels</code>.</p>
|
|
<p><code>bottleneck_channels</code> is less than <code>in_channels</code> and the $3x3$ convolution is performed
|
|
on this shrunk space (hence the bottleneck). The two $1x1$ convolution decreases and increases
|
|
the number of channels.</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">157</span><span class="k">class</span> <span class="nc">BottleneckResidualBlock</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-26'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-26'>#</a>
|
|
</div>
|
|
<ul>
|
|
<li><code>in_channels</code> is the number of channels in $x$</li>
|
|
<li><code>bottleneck_channels</code> is the number of channels for the $3x3$ convlution</li>
|
|
<li><code>out_channels</code> is the number of output channels</li>
|
|
<li><code>stride</code> is the stride length in the $3x3$ convolution operation.</li>
|
|
</ul>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">184</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">in_channels</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">bottleneck_channels</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">stride</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-27'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-27'>#</a>
|
|
</div>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">191</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-28'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-28'>#</a>
|
|
</div>
|
|
<p>First $1 \times 1$ convolution layer, this maps to <code>bottleneck_channels</code></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">194</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="n">in_channels</span><span class="p">,</span> <span class="n">bottleneck_channels</span><span class="p">,</span> <span class="n">kernel_size</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-29'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-29'>#</a>
|
|
</div>
|
|
<p>Batch normalization after the first convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">196</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="n">bottleneck_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-30'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-30'>#</a>
|
|
</div>
|
|
<p>First activation function (ReLU)</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">198</span> <span class="bp">self</span><span class="o">.</span><span class="n">act1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-31'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-31'>#</a>
|
|
</div>
|
|
<p>Second $3 \times 3$ convolution layer</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">201</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="n">bottleneck_channels</span><span class="p">,</span> <span class="n">bottleneck_channels</span><span class="p">,</span> <span class="n">kernel_size</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="n">stride</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-32'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-32'>#</a>
|
|
</div>
|
|
<p>Batch normalization after the second convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">203</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="n">bottleneck_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-33'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-33'>#</a>
|
|
</div>
|
|
<p>Second activation function (ReLU)</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">205</span> <span class="bp">self</span><span class="o">.</span><span class="n">act2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-34'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-34'>#</a>
|
|
</div>
|
|
<p>Third $1 \times 1$ convolution layer, this maps to <code>out_channels</code>.</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">208</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv3</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="n">bottleneck_channels</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">,</span> <span class="n">kernel_size</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-35'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-35'>#</a>
|
|
</div>
|
|
<p>Batch normalization after the second convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">210</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn3</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="n">out_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-36'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-36'>#</a>
|
|
</div>
|
|
<p>Shortcut connection should be a projection if the stride length is not $1$
|
|
of if the number of channels change</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">214</span> <span class="k">if</span> <span class="n">stride</span> <span class="o">!=</span> <span class="mi">1</span> <span class="ow">or</span> <span class="n">in_channels</span> <span class="o">!=</span> <span class="n">out_channels</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-37'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-37'>#</a>
|
|
</div>
|
|
<p>Projection $W_s x$</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">216</span> <span class="bp">self</span><span class="o">.</span><span class="n">shortcut</span> <span class="o">=</span> <span class="n">ShortcutProjection</span><span class="p">(</span><span class="n">in_channels</span><span class="p">,</span> <span class="n">out_channels</span><span class="p">,</span> <span class="n">stride</span><span class="p">)</span>
|
|
<span class="lineno">217</span> <span class="k">else</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-38'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-38'>#</a>
|
|
</div>
|
|
<p>Identity $x$</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">219</span> <span class="bp">self</span><span class="o">.</span><span class="n">shortcut</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Identity</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-39'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-39'>#</a>
|
|
</div>
|
|
<p>Second activation function (ReLU) (after adding the shortcut)</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">222</span> <span class="bp">self</span><span class="o">.</span><span class="n">act3</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-40'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-40'>#</a>
|
|
</div>
|
|
<ul>
|
|
<li><code>x</code> is the input of shape <code>[batch_size, in_channels, height, width]</code></li>
|
|
</ul>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">224</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-41'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-41'>#</a>
|
|
</div>
|
|
<p>Get the shortcut connection</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">229</span> <span class="n">shortcut</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">shortcut</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-42'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-42'>#</a>
|
|
</div>
|
|
<p>First convolution and activation</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">231</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">act1</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">bn1</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv1</span><span class="p">(</span><span class="n">x</span><span class="p">)))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-43'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-43'>#</a>
|
|
</div>
|
|
<p>Second convolution and activation</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">233</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">act2</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">bn2</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv2</span><span class="p">(</span><span class="n">x</span><span class="p">)))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-44'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-44'>#</a>
|
|
</div>
|
|
<p>Third convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">235</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn3</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv3</span><span class="p">(</span><span class="n">x</span><span class="p">))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-45'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-45'>#</a>
|
|
</div>
|
|
<p>Activation function after adding the shortcut</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">237</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">act3</span><span class="p">(</span><span class="n">x</span> <span class="o">+</span> <span class="n">shortcut</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-46'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-46'>#</a>
|
|
</div>
|
|
<h2>ResNet Model</h2>
|
|
<p>This is a the base of the resnet model without
|
|
the final linear layer and softmax for classification.</p>
|
|
<p>The resnet is made of stacked <a href="#residual_block">residual blocks</a> or
|
|
<a href="#bottleneck_residual_block">bottleneck residual blocks</a>.
|
|
The feature map size is halved after a few blocks with a block of stride length $2$.
|
|
The number of channels is increased when the feature map size is reduced.
|
|
Finally the feature map is average pooled to get a vector representation.</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">240</span><span class="k">class</span> <span class="nc">ResNetBase</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-47'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-47'>#</a>
|
|
</div>
|
|
<ul>
|
|
<li><code>n_blocks</code> is a list of of number of blocks for each feature map size.</li>
|
|
<li><code>n_channels</code> is the number of channels for each feature map size.</li>
|
|
<li><code>bottlenecks</code> is the number of channels the bottlenecks.
|
|
If this is <code>None</code>, <a href="#residual_block">residual blocks</a> are used.</li>
|
|
<li><code>img_channels</code> is the number of channels in the input.</li>
|
|
<li><code>first_kernel_size</code> is the kernel size of the initial convolution layer</li>
|
|
</ul>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">254</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">n_blocks</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="n">n_channels</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span>
|
|
<span class="lineno">255</span> <span class="n">bottlenecks</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="lineno">256</span> <span class="n">img_channels</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">3</span><span class="p">,</span> <span class="n">first_kernel_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">7</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-48'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-48'>#</a>
|
|
</div>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">265</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-49'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-49'>#</a>
|
|
</div>
|
|
<p>Number of blocks and number of channels for each feature map size</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">268</span> <span class="k">assert</span> <span class="nb">len</span><span class="p">(</span><span class="n">n_blocks</span><span class="p">)</span> <span class="o">==</span> <span class="nb">len</span><span class="p">(</span><span class="n">n_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-50'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-50'>#</a>
|
|
</div>
|
|
<p>If <a href="#bottleneck_residual_block">bottleneck residual blocks</a> are used,
|
|
the number of channels in bottlenecks should be provided for each feature map size</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">271</span> <span class="k">assert</span> <span class="n">bottlenecks</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="nb">len</span><span class="p">(</span><span class="n">bottlenecks</span><span class="p">)</span> <span class="o">==</span> <span class="nb">len</span><span class="p">(</span><span class="n">n_channels</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-51'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-51'>#</a>
|
|
</div>
|
|
<p>Initial convolution layer maps from <code>img_channels</code> to number of channels in the first
|
|
residual block (<code>n_channels[0]</code>)</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">275</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="n">img_channels</span><span class="p">,</span> <span class="n">n_channels</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span>
|
|
<span class="lineno">276</span> <span class="n">kernel_size</span><span class="o">=</span><span class="n">first_kernel_size</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">padding</span><span class="o">=</span><span class="n">first_kernel_size</span> <span class="o">//</span> <span class="mi">2</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-52'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-52'>#</a>
|
|
</div>
|
|
<p>Batch norm after initial convolution</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">278</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">(</span><span class="n">n_channels</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-53'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-53'>#</a>
|
|
</div>
|
|
<p>List of blocks</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">281</span> <span class="n">blocks</span> <span class="o">=</span> <span class="p">[]</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-54'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-54'>#</a>
|
|
</div>
|
|
<p>Number of channels from previous layer (or block)</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">283</span> <span class="n">prev_channels</span> <span class="o">=</span> <span class="n">n_channels</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-55'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-55'>#</a>
|
|
</div>
|
|
<p>Loop through each feature map size</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">285</span> <span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">channels</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">n_channels</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-56'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-56'>#</a>
|
|
</div>
|
|
<p>The first block for the new feature map size, will have a stride length of $2$
|
|
except fro the very first block</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">288</span> <span class="n">stride</span> <span class="o">=</span> <span class="mi">2</span> <span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">blocks</span><span class="p">)</span> <span class="o">==</span> <span class="mi">0</span> <span class="k">else</span> <span class="mi">1</span>
|
|
<span class="lineno">289</span>
|
|
<span class="lineno">290</span> <span class="k">if</span> <span class="n">bottlenecks</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-57'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-57'>#</a>
|
|
</div>
|
|
<p><a href="#residual_block">residual blocks</a> that maps from <code>prev_channels</code> to <code>channels</code></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">292</span> <span class="n">blocks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">ResidualBlock</span><span class="p">(</span><span class="n">prev_channels</span><span class="p">,</span> <span class="n">channels</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="n">stride</span><span class="p">))</span>
|
|
<span class="lineno">293</span> <span class="k">else</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-58'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-58'>#</a>
|
|
</div>
|
|
<p><a href="#bottleneck_residual_block">bottleneck residual blocks</a>
|
|
that maps from <code>prev_channels</code> to <code>channels</code></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">296</span> <span class="n">blocks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">BottleneckResidualBlock</span><span class="p">(</span><span class="n">prev_channels</span><span class="p">,</span> <span class="n">bottlenecks</span><span class="p">[</span><span class="n">i</span><span class="p">],</span> <span class="n">channels</span><span class="p">,</span>
|
|
<span class="lineno">297</span> <span class="n">stride</span><span class="o">=</span><span class="n">stride</span><span class="p">))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-59'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-59'>#</a>
|
|
</div>
|
|
<p>Change the number of channels</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">300</span> <span class="n">prev_channels</span> <span class="o">=</span> <span class="n">channels</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-60'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-60'>#</a>
|
|
</div>
|
|
<p>Add rest of the blocks - no change in feature map size or channels</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">302</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">n_blocks</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="o">-</span> <span class="mi">1</span><span class="p">):</span>
|
|
<span class="lineno">303</span> <span class="k">if</span> <span class="n">bottlenecks</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-61'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-61'>#</a>
|
|
</div>
|
|
<p><a href="#residual_block">residual blocks</a></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">305</span> <span class="n">blocks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">ResidualBlock</span><span class="p">(</span><span class="n">channels</span><span class="p">,</span> <span class="n">channels</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="mi">1</span><span class="p">))</span>
|
|
<span class="lineno">306</span> <span class="k">else</span><span class="p">:</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-62'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-62'>#</a>
|
|
</div>
|
|
<p><a href="#bottleneck_residual_block">bottleneck residual blocks</a></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">308</span> <span class="n">blocks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">BottleneckResidualBlock</span><span class="p">(</span><span class="n">channels</span><span class="p">,</span> <span class="n">bottlenecks</span><span class="p">[</span><span class="n">i</span><span class="p">],</span> <span class="n">channels</span><span class="p">,</span> <span class="n">stride</span><span class="o">=</span><span class="mi">1</span><span class="p">))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-63'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-63'>#</a>
|
|
</div>
|
|
<p>Stack the blocks</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">311</span> <span class="bp">self</span><span class="o">.</span><span class="n">blocks</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="o">*</span><span class="n">blocks</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-64'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-64'>#</a>
|
|
</div>
|
|
<ul>
|
|
<li><code>x</code> has shape <code>[batch_size, img_channels, height, width]</code></li>
|
|
</ul>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">313</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-65'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-65'>#</a>
|
|
</div>
|
|
<p>Initial convolution and batch normalization</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">319</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">bn</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv</span><span class="p">(</span><span class="n">x</span><span class="p">))</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-66'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-66'>#</a>
|
|
</div>
|
|
<p>Residual (or bottleneck) blocks</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">321</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">blocks</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-67'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-67'>#</a>
|
|
</div>
|
|
<p>Change <code>x</code> from shape <code>[batch_size, channels, h, w]</code> to <code>[batch_size, channels, h * w]</code></p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">323</span> <span class="n">x</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">x</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">x</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="o">-</span><span class="mi">1</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-68'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-68'>#</a>
|
|
</div>
|
|
<p>Global average pooling</p>
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">325</span> <span class="k">return</span> <span class="n">x</span><span class="o">.</span><span class="n">mean</span><span class="p">(</span><span class="n">dim</span><span class="o">=-</span><span class="mi">1</span><span class="p">)</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='footer'>
|
|
<a href="https://papers.labml.ai">Trending Research Papers</a>
|
|
<a href="https://labml.ai">labml.ai</a>
|
|
</div>
|
|
</div>
|
|
<script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.4/MathJax.js?config=TeX-AMS_HTML">
|
|
</script>
|
|
<!-- MathJax configuration -->
|
|
<script type="text/x-mathjax-config">
|
|
MathJax.Hub.Config({
|
|
tex2jax: {
|
|
inlineMath: [ ['$','$'] ],
|
|
displayMath: [ ['$$','$$'] ],
|
|
processEscapes: true,
|
|
processEnvironments: true
|
|
},
|
|
// Center justify equations in code and markdown cells. Elsewhere
|
|
// we use CSS to left justify single line equations in code cells.
|
|
displayAlign: 'center',
|
|
"HTML-CSS": { fonts: ["TeX"] }
|
|
});
|
|
|
|
</script>
|
|
<script>
|
|
function handleImages() {
|
|
var images = document.querySelectorAll('p>img')
|
|
|
|
console.log(images);
|
|
for (var i = 0; i < images.length; ++i) {
|
|
handleImage(images[i])
|
|
}
|
|
}
|
|
|
|
function handleImage(img) {
|
|
img.parentElement.style.textAlign = 'center'
|
|
|
|
var modal = document.createElement('div')
|
|
modal.id = 'modal'
|
|
|
|
var modalContent = document.createElement('div')
|
|
modal.appendChild(modalContent)
|
|
|
|
var modalImage = document.createElement('img')
|
|
modalContent.appendChild(modalImage)
|
|
|
|
var span = document.createElement('span')
|
|
span.classList.add('close')
|
|
span.textContent = 'x'
|
|
modal.appendChild(span)
|
|
|
|
img.onclick = function () {
|
|
console.log('clicked')
|
|
document.body.appendChild(modal)
|
|
modalImage.src = img.src
|
|
}
|
|
|
|
span.onclick = function () {
|
|
document.body.removeChild(modal)
|
|
}
|
|
}
|
|
|
|
handleImages()
|
|
</script>
|
|
</body>
|
|
</html> |