<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sergios Karagiannakos on Medium]]></title>
        <description><![CDATA[Stories by Sergios Karagiannakos on Medium]]></description>
        <link>https://medium.com/@SergiosKar?source=rss-1c90ee6cdd6a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*zMTTQyTEYtaAfa3GEJ9N0w.jpeg</url>
            <title>Stories by Sergios Karagiannakos on Medium</title>
            <link>https://medium.com/@SergiosKar?source=rss-1c90ee6cdd6a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 16:24:01 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@SergiosKar/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Build a Transformer in JAX from scratch: how to write and train your own models]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/build-a-transformer-in-jax-from-scratch-how-to-write-and-train-your-own-models-9aa02b5b28fd?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/1920/1*PS6z2GYsrSQ7rk-Z9hrWgQ.png" width="1920"></a></p><p class="medium-feed-snippet">How to develop and train a Transformer with JAX, Haiku and Optax. Learn by example how to code Deep Learning models in JAX</p><p class="medium-feed-link"><a href="https://medium.com/data-science/build-a-transformer-in-jax-from-scratch-how-to-write-and-train-your-own-models-9aa02b5b28fd?source=rss-1c90ee6cdd6a------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/build-a-transformer-in-jax-from-scratch-how-to-write-and-train-your-own-models-9aa02b5b28fd?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/9aa02b5b28fd</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ml-engineering]]></category>
            <category><![CDATA[jax]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Fri, 19 Mar 2021 00:00:00 GMT</pubDate>
            <atom:updated>2021-04-14T14:46:34.278Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Best Bootcamps and Programs to learn Machine Learning and Data Science]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/ai-summer/best-bootcamps-and-programs-to-learn-machine-learning-and-data-science-44698bf1c801?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/1920/0*8-SEzBgmt2F-2PCQ.jpg" width="1920"></a></p><p class="medium-feed-snippet">Bootcamps and interactive online platforms are an excellent way to acquire new skills. When it comes to Machine Learning and Data Science&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/ai-summer/best-bootcamps-and-programs-to-learn-machine-learning-and-data-science-44698bf1c801?source=rss-1c90ee6cdd6a------2">Continue reading on AI Summer »</a></p></div>]]></description>
            <link>https://medium.com/ai-summer/best-bootcamps-and-programs-to-learn-machine-learning-and-data-science-44698bf1c801?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/44698bf1c801</guid>
            <category><![CDATA[bootcamp]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[coding]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Sun, 24 Jan 2021 22:00:00 GMT</pubDate>
            <atom:updated>2021-02-19T06:37:45.924Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How to Use Data Augmentation to 10x Your Image Datasets]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/how-to-use-data-augmentation-to-10x-your-image-datasets-dab42858be55?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/1368/0*s57alo3rA_xYnBPT.jpg" width="1368"></a></p><p class="medium-feed-snippet">Data Augmentation for Images: Basic data manipulations, GANs and Meta-Learning</p><p class="medium-feed-link"><a href="https://medium.com/data-science/how-to-use-data-augmentation-to-10x-your-image-datasets-dab42858be55?source=rss-1c90ee6cdd6a------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/how-to-use-data-augmentation-to-10x-your-image-datasets-dab42858be55?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/dab42858be55</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[data-augmentation]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Mon, 06 Apr 2020 01:00:16 GMT</pubDate>
            <atom:updated>2020-04-26T02:24:20.855Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How to learn Deep Learning in 2020]]></title>
            <link>https://medium.com/ai-summer/how-to-learn-deep-learning-in-2020-d733d090ccbc?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/d733d090ccbc</guid>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Sun, 29 Mar 2020 00:00:00 GMT</pubDate>
            <atom:updated>2020-07-28T08:15:42.232Z</atom:updated>
            <content:encoded><![CDATA[<p>Are you looking for a place to learn Deep Learning? Whether you are a beginner or an experienced Machine Learning Engineer, I am sure you will find the below collection useful.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Xfikcq5Fzk3eceCX.jpg" /></figure><p><strong>In this post, we gathered together all of our resources and organized them in a step by step guide to help you learn all the popular Deep Learning architectures and algorithms as efficiently and as fast as possible.</strong></p><p>You will also find articles focused on specific applications such as Computer Vision and Natural Language Processing (NLP) but also posts about how Reinforcement Learning works.</p><p>So, without further ado, let’s get started…</p><h3>Deep Learning Architectures</h3><h4><a href="https://theaisummer.com/Neural_Network_from_scratch/">Neural network library from scratch</a></h4><p>In this post, you will build a Feedforward Neural Network from scratch using C++. You will implement the backpropagation algorithm, define the network’s structure and train it in GPU using OpenCL</p><h4><a href="https://theaisummer.com/Neural_Network_from_scratch_part2/">Convolutional Neural network library from scratch</a></h4><p>In the second part, you will extend the library by including Convolutional neural networks. You will define the convolutional and pooling layers and program the OpenCL kernels to run the backpropagation in parallel.</p><h4><a href="https://theaisummer.com/skip-connections/">Intuitive Explanation of Skip Connections in Deep Learning</a></h4><p>What are skip connections, why they solve the vanishing gradient problem and how they are applied to popular Convolutional Neural Network architectures such as ResNet, DenseNet and UNet.</p><h4><a href="https://theaisummer.com/Bitcon_prediction_LSTM/">Predict Bitcoin price with LSTM</a></h4><p>Learn how recurrent neural networks works, what are LSTMs and what problem do they solve and finally use one to predict the bitcoin price using Python and Keras</p><h4><a href="https://theaisummer.com/Autoencoder/">How to Generate Images using Autoencoders</a></h4><p>The inner working of Autoencoders, what is a Variational Autoencoder (VAE) and how they can be used to generate original new images using PyTorch</p><h4><a href="https://theaisummer.com/Generative_Artificial_Intelligence/">Decrypt Generative Artificial Intelligence and GANs</a></h4><p>How Generative models differ from other machine learning architectures, how Generative Adversarial Networks (GAN) learn from data and why they are able to generate new data points?</p><h3><a href="https://theaisummer.com/Graph_Neural_Networks/">Graph Neural Networks</a></h3><p>Neural Networks can also be applied in graph data besides images and text. Graph Neural Networks have the ability to take a Graph as an input and encode its information into a single numeric vector.</p><h3><a href="https://theaisummer.com/NALU/">Explain Neural Arithmetic Logic Units (NALU)</a></h3><p>Neural Arithmetic Logic Units solve a problem that most machine learning architectures can’t tackle. They are able to count. With NALU we can perform arithmetic operations such additions and multiplications and approximate simple arithmetic functions.</p><h3>Computer Vision and Deep Learning</h3><h4><a href="https://theaisummer.com/Semantic_Segmentation/">Semantic Segmentation in the era of Neural Networks</a></h4><p>In semantic segmentation, the goal is to classify each pixel of the image in a specific category based. That way we can extract contextual information of every object in the image. To achieve this, skip connections are used in neural networks, forming a new architecture called UNets.</p><p>Understand how UNets work, why the perform well in semantic segmentation and program one using Keras.</p><h4><a href="https://theaisummer.com/Localization_and_Object_Detection/">Localization and Object Detection with Deep Learning</a></h4><p>Localization is the task of identifying the location of an object in an image, while Object Detection is the classification and detection of all objects in it. To do this, the most popular method is an R-CNN alongside with its improvements Fast R-CNN and Faster R-CNN.</p><h4><a href="https://theaisummer.com/YOLO/">YOLO — You only look once</a></h4><p>SIngle shot detectors like YOLO provide a fast way to detect and localize objects in an image. In this post, you will learn the secrets behind YOLO and why it became the industry standard in low-power devices such as smartphones.</p><h3>Applications</h3><h4><a href="https://theaisummer.com/Self_driving_cars/">Self-driving cars using Deep Learning</a></h4><p>Learn the basics steps behind the development of a car’s autopilots and use a game simulator and python to make your own car drive all by itself.</p><h4><a href="https://theaisummer.com/Human-Pose-Estimation/">Human Pose Estimation</a></h4><p>This is an overview of the most important research papers on 2D and 3D Human Pose Estimation. You will find intuitive explanations on algorithms like OpenPose, DensePose and VIBE.</p><h4><a href="https://theaisummer.com/medical-image-deep-learning/">Deep learning in medical imaging: 3D medical image segmentation with PyTorch</a></h4><p>In this document, we tackle the 3D medical image segmentation with deep learning models using PyTorch. The basic MRI foundations are presented for tensor representation, as well as the basic components to apply a deep learning method that handles the task-specific problems(class imbalance, limited data). Moreover, we present some features of the open source medical image segmentation library. Finally, we discuss our preliminary experimental results and provide sources to find medical imaging data.</p><h3>Reinforcement Learning</h3><h4><a href="https://theaisummer.com/Reinforcement_learning/">The secrets behind Reinforcement Learning</a></h4><p>The purpose of this article is to give you an idea of the fundamentals principles behind reinforcement learning, how we define our agent, the environment and the learning process. It also includes an overview of the different types of RL algorithms and the basic logic behind them.</p><h4><a href="https://theaisummer.com/Deep_Q_Learning/">Deep Q Learning</a></h4><p>Here we dive into Q Learning, we analyze what exactly the Q value is and how we can approximate it and also how neural networks and deep learning revolutionize this technique. You will also find code examples on how to build your own Deep Q Learning agent in Python.</p><h4><a href="https://theaisummer.com/Taking_Deep_Q_Networks_a_step_further/">Taking Deep Q Networks a step further</a></h4><p>As a continuance of the previous post, here we introduce topics like Moving and Fixed Q targets, concepts like Maximation Bias and Experience Replay and we describe how Double Deep Q Networks and Dueling Deep Q Networks improve over the original idea</p><h4><a href="https://theaisummer.com/Policy-Gradients/">Unravel Policy Gradients and REINFORCE</a></h4><p>What are policy-based methods, why they are different from value-based methods and what’s the idea behind Monte Carlo policy gradients (aka REINFORCE). As always python code is available in the end.</p><h4><a href="https://theaisummer.com/Actor_critics/">The idea behind Actor-Critics and how A2C and A3C improve them</a></h4><p>In this article, you will learn about actor critics and many variations of them such as Advantage Actor-Critics (A2C) and Asynchronous Advantage Actor-Critics (A3C)</p><h4><a href="https://theaisummer.com/TRPO_PPO/">Trust Region and Proximal policy optimization (TRPO and PPO)</a></h4><p>Returning to policy methods, we present two of the most recent algorithms in the field: Trust region policy optimization (TRPO) and Proximal policy optimization (PPO)</p><h3>Conclusion</h3><p>In this list, we tried to cover the most important areas of Deep Learning in a structured way so you can start from zero. All these articles are a product of years of research and practice. But for sure it doesn’t even begin to cover all the existent topics out there. There is so much more yet to come.</p><p>If you want to follow us in our journey towards solving AI and become an expert in Deep Learning, don’t forget to subscribe to our <a href="https://theaisummer.com/newsletter/">newsletter</a>.</p><p>The AI Summer team</p><p><em>Originally published at </em><a href="https://theaisummer.com/learn-deep-learning/"><em>https://theaisummer.com</em></a><em> on March 29, 2020.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d733d090ccbc" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ai-summer/how-to-learn-deep-learning-in-2020-d733d090ccbc">How to learn Deep Learning in 2020</a> was originally published in <a href="https://medium.com/ai-summer">AI Summer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deep Learning Algorithms — The Complete Guide]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/deep-learning-algorithms-the-complete-guide-e4e7c535b2fc?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/1920/1*2ecAbaeLoHmVU2cf3A44zA.png" width="1920"></a></p><p class="medium-feed-snippet">The most popular deep learning algorithms including those used in Computer Vision and NLP</p><p class="medium-feed-link"><a href="https://medium.com/data-science/deep-learning-algorithms-the-complete-guide-e4e7c535b2fc?source=rss-1c90ee6cdd6a------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/deep-learning-algorithms-the-complete-guide-e4e7c535b2fc?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/e4e7c535b2fc</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[naturallanguageprocessing]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Wed, 26 Feb 2020 00:00:00 GMT</pubDate>
            <atom:updated>2020-03-11T17:42:47.564Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Best Artificial Intelligence books to read]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/best-artificial-intelligence-books-to-read-30d19ece0799?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/2600/1*NKcpANrUtUznH-nzmxPM3Q.jpeg" width="3136"></a></p><p class="medium-feed-snippet">The ultimate list of Artificial Intelligence and Machine Learning non-fiction books</p><p class="medium-feed-link"><a href="https://medium.com/data-science/best-artificial-intelligence-books-to-read-30d19ece0799?source=rss-1c90ee6cdd6a------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/best-artificial-intelligence-books-to-read-30d19ece0799?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/30d19ece0799</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[superintelligence]]></category>
            <category><![CDATA[books]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Sat, 15 Feb 2020 00:00:00 GMT</pubDate>
            <atom:updated>2020-03-04T15:53:33.706Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Top 10 courses to learn Machine and Deep Learning (2020)]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/top-10-courses-to-learn-machine-and-deep-learning-2020-7cc2d9c3bae3?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/1920/0*kr4MUojaHt-zaoe6.jpg" width="1920"></a></p><p class="medium-feed-snippet">Machine Learning Courses&#x200A;&#x2014;&#x200A;The ultimate list</p><p class="medium-feed-link"><a href="https://medium.com/data-science/top-10-courses-to-learn-machine-and-deep-learning-2020-7cc2d9c3bae3?source=rss-1c90ee6cdd6a------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/top-10-courses-to-learn-machine-and-deep-learning-2020-7cc2d9c3bae3?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/7cc2d9c3bae3</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[online-courses]]></category>
            <category><![CDATA[course]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Mon, 10 Feb 2020 00:00:00 GMT</pubDate>
            <atom:updated>2020-02-19T15:14:47.466Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How to get hired as a Machine Learning Engineer]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/how-to-get-hired-as-a-machine-learning-engineer-a9d4dca34dbf?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/1915/0*-OtTxJcSTdOCCw7C.jpg" width="1915"></a></p><p class="medium-feed-snippet">7 steps to land a Machine Learning Engineer Job</p><p class="medium-feed-link"><a href="https://medium.com/data-science/how-to-get-hired-as-a-machine-learning-engineer-a9d4dca34dbf?source=rss-1c90ee6cdd6a------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/how-to-get-hired-as-a-machine-learning-engineer-a9d4dca34dbf?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/a9d4dca34dbf</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[coding-interviews]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[job-hunting]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Sat, 08 Feb 2020 00:00:00 GMT</pubDate>
            <atom:updated>2020-02-15T16:58:42.350Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Graph Neural Networks — An overview]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/graph-neural-networks-an-overview-dfd363b6ef87?source=rss-1c90ee6cdd6a------2"><img src="https://cdn-images-1.medium.com/max/700/0*7ykfn2_nudrPJGRZ.jpg" width="700"></a></p><p class="medium-feed-snippet">Over the past decade, we&#x2019;ve seen that Neural Networks can perform tremendously well in structured data like images and text. Most of the&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/data-science/graph-neural-networks-an-overview-dfd363b6ef87?source=rss-1c90ee6cdd6a------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/graph-neural-networks-an-overview-dfd363b6ef87?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/dfd363b6ef87</guid>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[graph]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Sat, 01 Feb 2020 00:00:00 GMT</pubDate>
            <atom:updated>2020-02-18T18:27:04.692Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[YOLO — You only look once]]></title>
            <link>https://medium.com/data-science/yolo-you-only-look-once-c63247e59d66?source=rss-1c90ee6cdd6a------2</link>
            <guid isPermaLink="false">https://medium.com/p/c63247e59d66</guid>
            <category><![CDATA[localization]]></category>
            <category><![CDATA[object-detection]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Sergios Karagiannakos]]></dc:creator>
            <pubDate>Mon, 13 May 2019 00:00:00 GMT</pubDate>
            <atom:updated>2019-05-13T13:17:10.977Z</atom:updated>
            <content:encoded><![CDATA[<h3>YOLO — You only look once</h3><h4>YOLO!!! So do we only live once? I sure do not know. What I know is that we only have to LOOK once. Wait what?</h4><p>That’s right. If you want to detect and localize objects in an image, there is no need to go through the whole process of proposing regions of interest, classify them and correct their bounding boxes. If you recall from my previous <a href="https://sergioskar.github.io/Localization_and_Object_Detection/">article</a>, this is exactly what models like RCNN and Faster RCNN do.</p><p>Do we really need all that complexity and computation? Well if we want top-notch accuracy we certainly do. Luckily there is another simpler way to perform such a task, by processing the image only once and output the prediction immediately. These types of models are called Single shot detectors.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yy-P07HSJSRb1_4a.jpg" /><figcaption><a href="https://www.youtube.com/watch?v=yQwfDxBMtXg">https://www.youtube.com/watch?v=yQwfDxBMtXg</a></figcaption></figure><h3>Single shot detectors</h3><p>Instead of having a dedicated system to propose regions of interests, we have a set of predefined boxes to look for objects, which are forwarded to a bunch of convolutional layers to predict class scores and bounding box offsets. <strong>Then for each box we predict a number of bounding boxes with a confidence score assigned to each one, we detect one object centered in that box and we output a set of probabilities for each possible class</strong>. Once we have all that, we simply and maybe naively keep only the box with a high confidence score. And it works. With very impressive results actually. To elaborate the overall flow even better, let’s use one of the most popular single shot detectors called YOLO .</p><h3>You only look once (YOLO)</h3><p>There have been 3 versions of the model so far, with each new one improving the previous in terms of both speed and accuracy. The number of predefined cells and the number of predicted bounding boxes for each cell is defined based on the input size and the classes. In our case, we are going to use the actual numbers used to evaluate the <a href="http://host.robots.ox.ac.uk/pascal/VOC/">PASCAL VOC</a> dataset.</p><p>First, we divide the image into a grid of 13x13, resulting in 169 cells in total.</p><p>For every one of the cells, it predicts 5 bounding boxes (x,y,w,h) with a confidence score, it detects one object regardless the number of boxes and 20 probabilities for the 20 classes.</p><p>In total, we have 169*5=845 bounding boxes and the shape of output tensor of the mode is going to be (13,13,5*5+20)= (13,13,45). The whole essence of the YOLO models is to build this (13,13,45) tensor. To accomplish that, it uses a CNN network and 2 fully connected layers to perform the actual regression.</p><p>The final prediction is extracted after keeping only the bounding boxes with a high confidence score( higher than a threshold such as 0.3)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*4I1CRBHhfocoVaSa.jpg" /><figcaption><a href="https://pjreddie.com/darknet/yolo/"><em>https://pjreddie.com/darknet/yolo/</em></a></figcaption></figure><p>Because the model may output duplicate detections for the same object, we use a technique called <strong>Non-maximal suppression</strong> to remove duplicates. In a simple implementation, we sort the predictions by the confidence score and as we iterate them we keep only the first appearances of each class.</p><p>As far as the actual model is concerned, the architecture is quite trivial as it consists of only convolutional and pooling layers, without any fancy tricks. We train the model using a multiple loss function, which includes a classification loss, a localization loss and a confidence loss.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JGGHFj08Pp2HQ-PL.jpg" /></figure><p>The most recent versions of YOLO have introduced some special tricks to improve the accuracy and reduce the training and inference time. Some examples are batch normalization, anchor boxes, dimensions clusters and others. If you want to get into more details, you should definitely check the original <a href="https://pjreddie.com/publications/">papers</a>.</p><p>Also to dive into code and try the YOLO models in practice, check out these two awesome repositories in Github (<a href="https://github.com/qqwweee/keras-yolo3">repo1</a> and <a href="https://github.com/experiencor/keras-yolo2">repo2</a>).</p><p>The power of YOLO is not its spectacular accuracy or the very clever ideas behind it, is its superb speed, which makes it ideal for embedded systems and low-power applications. That’s why self-driving cars and surveillance cameras are its most common real-world use cases.</p><p>As deep learning continues to play along with computer vision (and it will sure do), we can expect many more models to be tailored for low-power systems even if they sometimes sacrifice accuracy. And don’t forget the whole Internet of Things kind of thing. This is where these models really shine.</p><blockquote><strong>If you have any thoughts, comments, questions or you just want to stay updated on my latest content, feel free to connect with me on </strong><a href="https://www.linkedin.com/in/sergios-karagiannakos/"><strong><em>Linkedin</em></strong></a><strong><em>, </em></strong><a href="https://twitter.com/KarSergios"><strong><em>Twitter</em></strong></a><strong><em>, </em></strong><a href="https://www.instagram.com/sergios_krg/"><strong><em>Instagram</em></strong></a><strong><em>, </em></strong><a href="https://github.com/SergiosKar"><strong><em>Github</em></strong></a><strong><em>, or on my </em></strong><a href="https://sergioskar.github.io/"><strong><em>Website</em></strong></a><strong><em>.</em></strong></blockquote><p><em>Originally published at </em><a href="https://sergioskar.github.io/YOLO/"><em>https://sergioskar.github.io</em></a><em> on May 13, 2019.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c63247e59d66" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/yolo-you-only-look-once-c63247e59d66">YOLO — You only look once</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>