<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>Deep-Learning on Daniel Mejía R.</title>
		<link>https://daniel-m.github.io/tags/deep-learning/</link>
		<description>Recent content in Deep-Learning on Daniel Mejía R.</description>
		<generator>Hugo</generator>
		<language>en</language>
		
		
		
		
			<lastBuildDate>Thu, 19 Nov 2026 19:28:36 -0500</lastBuildDate>
		
			<atom:link href="https://daniel-m.github.io/tags/deep-learning/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>ImageNet: VGGNet, ResNet, Inception, and Xception with Keras</title>
				<link>https://daniel-m.github.io/2018/06/imagenet-vggnet-resnet-inception-and-xception-with-keras/</link>
				<pubDate>Fri, 01 Jun 2018 21:47:48 -0500</pubDate>
				<guid>https://daniel-m.github.io/2018/06/imagenet-vggnet-resnet-inception-and-xception-with-keras/</guid>
				<description>&lt;h4 id=&#34;link&#34;&gt;&#xA;  Link: &lt;a href=&#34;https://www.pyimagesearch.com/2017/03/20/imagenet-vggnet-resnet-inception-xception-keras/&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://www.pyimagesearch.com/2017/03/20/imagenet-vggnet-resnet-inception-xception-keras/&lt;/a&gt;&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#link&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h4&gt;&#xA;&lt;h2 id=&#34;key-concepts&#34;&gt;&#xA;  Key concepts:&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#key-concepts&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h2&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;CNN: Convolutional Neural Network&lt;/li&gt;&#xA;&lt;li&gt;ImageNet dataset&lt;/li&gt;&#xA;&lt;li&gt;ImageNet Large Scale Visual Recognition Challenge:  train a model that can correctly classify an input image into 1,000 separate object categories. ImageNet challenge is the de facto benchmark for computer vision classification algorithms&lt;/li&gt;&#xA;&lt;li&gt;Weight serialization&lt;/li&gt;&#xA;&lt;li&gt;Weights of a neural network&lt;/li&gt;&#xA;&lt;li&gt;Pooling&lt;/li&gt;&#xA;&lt;li&gt;Softmax&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;h2 id=&#34;notes&#34;&gt;&#xA;  Notes:&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#notes&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h2&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;Keras comes integrated with VGG16, VGG19, ResNet50, Inception V3, and Xception neural network models (look inside applications submodules)&lt;/li&gt;&#xA;&lt;li&gt;ImageNet: manually labeled 22 000 object categories&lt;/li&gt;&#xA;&lt;li&gt;ImageNet Large Scale Visual Recognition Challenge:  train a model that can correctly classify an input image into 1,000 separate object categories.&lt;/li&gt;&#xA;&lt;li&gt;ImageNet challenge is the de facto benchmark for computer vision classification algorithms&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;h3 id=&#34;vgg16-vgg19&#34;&gt;&#xA;  VGG16 VGG19&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#vgg16-vgg19&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h3&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;arXiv:&#xA;&lt;ul&gt;&#xA;&lt;li&gt;&lt;a href=&#34;https://arxiv.org/abs/1409.1556&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://arxiv.org/abs/1409.1556&lt;/a&gt;&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;The VGG network architecture was introduced by Simonyan and Zisserman in their 2014 paper, Very Deep Convolutional Networks for Large Scale Image Recognition.&lt;/li&gt;&#xA;&lt;li&gt;3×3 convolutional layers stacked on top of each other in increasing depth&lt;/li&gt;&#xA;&lt;li&gt;The “16” and “19” stand for the number of weight layers in the network&lt;/li&gt;&#xA;&lt;li&gt;The smaller networks converged and were then used as initializations for the larger, deeper networks — this process is called pre-training.&lt;/li&gt;&#xA;&lt;li&gt;Slow to train&lt;/li&gt;&#xA;&lt;li&gt;Huge compared to other networks, in terms of bandwidth and disk usage&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;h3 id=&#34;resnet&#34;&gt;&#xA;  ResNet&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#resnet&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h3&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;arXiv:&#xA;&lt;ul&gt;&#xA;&lt;li&gt;&lt;a href=&#34;https://arxiv.org/abs/1512.03385&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://arxiv.org/abs/1512.03385&lt;/a&gt;&lt;/li&gt;&#xA;&lt;li&gt;&lt;a href=&#34;https://arxiv.org/abs/1603.05027&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://arxiv.org/abs/1603.05027&lt;/a&gt;&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;First introduced by He et al. in their 2015 paper, Deep Residual Learning for Image Recognition&lt;/li&gt;&#xA;&lt;li&gt;Micro-architecture modules -&amp;gt; building blocks to construct the network&lt;/li&gt;&#xA;&lt;li&gt;ResNet50 -&amp;gt; 50 weight layers, based on 2015 paper&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;h3 id=&#34;inception-v3&#34;&gt;&#xA;  Inception V3&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#inception-v3&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h3&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;arXiv:&#xA;&lt;ul&gt;&#xA;&lt;li&gt;&lt;a href=&#34;https://arxiv.org/abs/1409.4842&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://arxiv.org/abs/1409.4842&lt;/a&gt;&lt;/li&gt;&#xA;&lt;li&gt;&lt;a href=&#34;https://arxiv.org/abs/1512.00567&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://arxiv.org/abs/1512.00567&lt;/a&gt;&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;First introduced by Szegedy et al. in their 2014 paper, Going Deeper with Convolutions&lt;/li&gt;&#xA;&lt;li&gt;The goal of the inception module is to act as a “multi-level feature extractor” by computing 1×1, 3×3, and 5×5 convolutions within the same module of the network&lt;/li&gt;&#xA;&lt;li&gt;GoogLeNet -&amp;gt; Inception vN, N = version number put by google&lt;/li&gt;&#xA;&lt;li&gt;Keras has inception from paper of 2015&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;h3 id=&#34;xception&#34;&gt;&#xA;  Xception&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#xception&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h3&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;arXiv:&#xA;&lt;ul&gt;&#xA;&lt;li&gt;&lt;a href=&#34;https://arxiv.org/abs/1610.02357&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://arxiv.org/abs/1610.02357&lt;/a&gt;&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;The original publication, Xception: Deep Learning with Depthwise Separable Convolutions, by  François Chollet&lt;/li&gt;&#xA;&lt;li&gt;Smallest weight serialization&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;h3 id=&#34;squeezenet&#34;&gt;&#xA;  SqueezeNet&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#squeezenet&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h3&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;arXiv:&#xA;&lt;ul&gt;&#xA;&lt;li&gt;&lt;a href=&#34;https://arxiv.org/abs/1602.07360&#34;  class=&#34;external-link&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;https://arxiv.org/abs/1602.07360&lt;/a&gt;&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;Can achieve levels of accuracy similar to AlexNet.&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;hr&gt;&#xA;&lt;h2 id=&#34;classifying-images-with-vggnet-resnet-inception-and-xception-with-python-and-keras&#34;&gt;&#xA;  Classifying images with VGGNet, ResNet, Inception, and Xception with Python and Keras&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#classifying-images-with-vggnet-resnet-inception-and-xception-with-python-and-keras&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h2&gt;&#xA;&lt;h3 id=&#34;cnn-image-pixel-sizes&#34;&gt;&#xA;  CNN Image pixel sizes&#xA;  &lt;a class=&#34;heading-link&#34; href=&#34;#cnn-image-pixel-sizes&#34;&gt;&#xA;    &lt;i class=&#34;fa-solid fa-link&#34; aria-hidden=&#34;true&#34; title=&#34;Link to heading&#34;&gt;&lt;/i&gt;&#xA;    &lt;span class=&#34;sr-only&#34;&gt;Link to heading&lt;/span&gt;&#xA;  &lt;/a&gt;&#xA;&lt;/h3&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;ImageNet : 224×224, 227×227, 256×256, and 299×299; however, you may see other dimensions as well.&lt;/li&gt;&#xA;&lt;li&gt;VGG16 : 224×224&lt;/li&gt;&#xA;&lt;li&gt;VGG19 : 224×224&lt;/li&gt;&#xA;&lt;li&gt;ResNet : 224×224&lt;/li&gt;&#xA;&lt;li&gt;Inception V3: 299x299&lt;/li&gt;&#xA;&lt;li&gt;Xception: 299×299 pixel inputs&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;p&gt;The weights for the chosen neural network will be downloaded in the first use and will be available for later uses with Keras&lt;/p&gt;</description>
			</item>
	</channel>
</rss>
