book

Deep Learning for Vision Systems

Name: Deep Learning for Vision Systems
Author: Mohamed Elgendy
ISBN: 9781617296192

by Mohamed Elgendy

November 2020

Intermediate to advanced

480 pages

13h 56m

English

Manning Publications

Audiobook available

Read now

Unlock full access

Deep Learning for Vision Systems
Copyright
dedication
contents
front matter
prefaceacknowledgmentsabout this bookWho should read this bookHow this book is organized: A roadmapAbout the codeliveBook discussion forumabout the authorabout the cover illustration
Part 1. Deep learning foundation
1 Welcome to computer vision
1.1 Computer vision1.1.1 What is visual perception?1.1.2 Vision systems1.1.3 Sensing devices1.1.4 Interpreting devices1.2 Applications of computer vision1.2.1 Image classification1.2.2 Object detection and localization1.2.3 Generating art (style transfer)1.2.4 Creating images1.2.5 Face recognition1.2.6 Image recommendation system1.3 Computer vision pipeline: The big picture1.4 Image input1.4.1 Image as functions1.4.2 How computers see images1.4.3 Color images1.5 Image preprocessing1.5.1 Converting color images to grayscale to reduce computation complexity1.6 Feature extraction1.6.1 What is a feature in computer vision?1.6.2 What makes a good (useful) feature?1.6.3 Extracting features (handcrafted vs. automatic extracting)1.7 Classifier learning algorithmSummary
2 Deep learning and neural networks
2.1 Understanding perceptrons2.1.1 What is a perceptron?2.1.2 How does the perceptron learn?2.1.3 Is one neuron enough to solve complex problems?2.2 Multilayer perceptrons2.2.1 Multilayer perceptron architecture2.2.2 What are hidden layers?2.2.3 How many layers, and how many nodes in each layer?2.2.4 Some takeaways from this section2.3 Activation functions2.3.1 Linear transfer function2.3.2 Heaviside step function (binary classifier)2.3.3 Sigmoid/logistic functionSoftmax function2.3.5 Hyperbolic tangent function (tanh)2.3.6 Rectified linear unit2.3.7 Leaky ReLU2.4 The feedforward process2.4.1 Feedforward calculations2.4.2 Feature learning2.5 Error functions2.5.1 What is the error function?2.5.2 Why do we need an error function?2.5.3 Error is always positive2.5.4 Mean square error2.5.5 Cross-entropy2.5.6 A final note on errors and weights2.6 Optimization algorithms2.6.1 What is optimization?2.6.2 Batch gradient descent2.6.3 Stochastic gradient descent2.6.4 Mini-batch gradient descent2.6.5 Gradient descent takeaways2.7 Backpropagation2.7.1 What is backpropagation?2.7.2 Backpropagation takeawaysSummary
3 Convolutional neural networks
3.1 Image classification using MLP3.1.1 Input layer3.1.2 Hidden layers3.1.3 Output layer3.1.4 Putting it all together3.1.5 Drawbacks of MLPs for processing images3.2 CNN architecture3.2.1 The big picture3.2.2 A closer look at feature extraction3.2.3 A closer look at classification3.3 Basic components of a CNN3.3.1 Convolutional layers3.3.2 Pooling layers or subsampling3.3.3 Fully connected layers3.4 Image classification using CNNs3.4.1 Building the model architecture3.4.2 Number of parameters (weights)3.5 Adding dropout layers to avoid overfitting3.5.1 What is overfitting?3.5.2 What is a dropout layer?3.5.3 Why do we need dropout layers?3.5.4 Where does the dropout layer go in the CNN architecture?3.6 Convolution over color images (3D images)3.6.1 How do we perform a convolution on a color image?3.6.2 What happens to the computational complexity?3.7 Project: Image classification for color imagesSummary
4 Structuring DL projects and hyperparameter tuning
4.1 Defining performance metrics4.1.1 Is accuracy the best metric for evaluating a model?4.1.2 Confusion matrix4.1.3 Precision and recall4.1.4 F-score4.2 Designing a baseline model4.3 Getting your data ready for training4.3.1 Splitting your data for train/validation/test4.3.2 Data preprocessing4.4 Evaluating the model and interpreting its performance4.4.1 Diagnosing overfitting and underfitting4.4.2 Plotting the learning curves4.4.3 Exercise: Building, training, and evaluating a network4.5 Improving the network and tuning hyperparameters4.5.1 Collecting more data vs. tuning hyperparameters4.5.2 Parameters vs. hyperparameters4.5.3 Neural network hyperparameters4.5.4 Network architecture4.6 Learning and optimization4.6.1 Learning rate and decay schedule4.6.2 A systematic approach to find the optimal learning rate4.6.3 Learning rate decay and adaptive learning4.6.4 Mini-batch size4.7 Optimization algorithms4.7.1 Gradient descent with momentum4.7.2 Adam4.7.3 Number of epochs and early stopping criteria4.7.4 Early stopping4.8 Regularization techniques to avoid overfitting4.8.1 L2 regularization4.8.2 Dropout layers4.8.3 Data augmentation4.9 Batch normalization4.9.1 The covariate shift problem4.9.2 Covariate shift in neural networks4.9.3 How does batch normalization work?4.9.4 Batch normalization implementation in Keras4.9.5 Batch normalization recap4.10 Project: Achieve high accuracy on image classificationSummary

Part 2. Image classification and detection
5 Advanced CNN architectures
5.1 CNN design patterns5.2 LeNet-55.2.1 LeNet architecture5.2.2 LeNet-5 implementation in Keras5.2.3 Setting up the learning hyperparameters5.2.4 LeNet performance on the MNIST dataset5.3 AlexNet5.3.1 AlexNet architecture5.3.2 Novel features of AlexNet5.3.3 AlexNet implementation in Keras5.3.4 Setting up the learning hyperparameters5.3.5 AlexNet performance5.4 VGGNet5.4.1 Novel features of VGGNet5.4.2 VGGNet configurations5.4.3 Learning hyperparameters5.4.4 VGGNet performance5.5 Inception and GoogLeNet5.5.1 Novel features of Inception5.5.2 Inception module: Naive version5.5.3 Inception module with dimensionality reduction5.5.4 Inception architecture5.5.5 GoogLeNet in Keras5.5.6 Learning hyperparameters5.5.7 Inception performance on the CIFAR dataset5.6 ResNet5.6.1 Novel features of ResNet5.6.2 Residual blocks5.6.3 ResNet implementation in Keras5.6.4 Learning hyperparameters5.6.5 ResNet performance on the CIFAR datasetSummary
6 Transfer learning
6.1 What problems does transfer learning solve?6.2 What is transfer learning?6.3 How transfer learning works6.3.1 How do neural networks learn features?6.3.2 Transferability of features extracted at later layers6.4 Transfer learning approaches6.4.1 Using a pretrained network as a classifier6.4.2 Using a pretrained network as a feature extractor6.4.3 Fine-tuning6.5 Choosing the appropriate level of transfer learning6.5.1 Scenario 1: Target dataset is small and similar to the source dataset6.5.2 Scenario 2: Target dataset is large and similar to the source dataset6.5.3 Scenario 3: Target dataset is small and different from the source dataset6.5.4 Scenario 4: Target dataset is large and different from the source dataset6.5.5 Recap of the transfer learning scenarios6.6 Open source datasets6.6.1 MNIST6.6.2 Fashion-MNIST6.6.3 CIFAR6.6.4 ImageNet6.6.5 MS COCO6.6.6 Google Open Images6.6.7 Kaggle6.7 Project 1: A pretrained network as a feature extractor6.8 Project 2: Fine-tuningSummary
7 Object detection with R-CNN, SSD, and YOLO
7.1 General object detection framework7.1.1 Region proposals7.1.2 Network predictions7.1.3 Non-maximum suppression (NMS)7.1.4 Object-detector evaluation metrics7.2 Region-based convolutional neural networks (R-CNNs)7.2.1 R-CNN7.2.2 Fast R-CNN7.2.3 Faster R-CNN7.2.4 Recap of the R-CNN family7.3 Single-shot detector (SSD)7.3.1 High-level SSD architecture7.3.2 Base network7.3.3 Multi-scale feature layers7.3.4 Non-maximum suppression7.4 You only look once (YOLO)7.4.1 How YOLOv3 works7.4.2 YOLOv3 architecture7.5 Project: Train an SSD network in a self-driving car application7.5.1 Step 1: Build the model7.5.2 Step 2: Model configuration7.5.3 Step 3: Create the model7.5.4 Step 4: Load the data7.5.5 Step 5: Train the model7.5.6 Step 6: Visualize the loss7.5.7 Step 7: Make predictionsSummary
Part 3. Generative models and visual embeddings
8 Generative adversarial networks (GANs)
8.1 GAN architecture8.1.1 Deep convolutional GANs (DCGANs)8.1.2 The discriminator model8.1.3 The generator model8.1.4 Training the GAN8.1.5 GAN minimax function8.2 Evaluating GAN models8.2.1 Inception score8.2.2 Fréchet inception distance (FID)8.2.3 Which evaluation scheme to use8.3 Popular GAN applications8.3.1 Text-to-photo synthesis8.3.2 Image-to-image translation (Pix2Pix GAN)8.3.3 Image super-resolution GAN (SRGAN)8.3.4 Ready to get your hands dirty?8.4 Project: Building your own GANSummary
9 DeepDream and neural style transfer
9.1 How convolutional neural networks see the world9.1.1 Revisiting how neural networks work9.1.2 Visualizing CNN features9.1.3 Implementing a feature visualizer9.2 DeepDream9.2.1 How the DeepDream algorithm works9.2.2 DeepDream implementation in Keras9.3 Neural style transfer9.3.1 Content loss9.3.2 Style loss9.3.3 Total variance loss9.3.4 Network trainingSummary
10 Visual embeddings
10.1 Applications of visual embeddings10.1.1 Face recognition10.1.2 Image recommendation systems10.1.3 Object re-identification10.2 Learning embedding10.3 Loss functions10.3.1 Problem setup and formalization10.3.2 Cross-entropy loss10.3.3 Contrastive loss10.3.4 Triplet loss10.3.5 Naive implementation and runtime analysis of losses10.4 Mining informative data10.4.1 Dataloader10.4.2 Informative data mining: Finding useful triplets10.4.3 Batch all (BA)10.4.4 Batch hard (BH)10.4.5 Batch weighted (BW)10.4.6 Batch sample (BS)10.5 Project: Train an embedding network10.5.1 Fashion: Get me items similar to this10.5.2 Vehicle re-identification10.5.3 Implementation10.5.4 Testing a trained model10.6 Pushing the boundaries of current accuracySummaryReferences
appendix A. Getting set up
A.1 Downloading the code repositoryA.2 Installing AnacondaA.3 Setting up your DL environmentA.3.1 Setting up your development environment manuallyA.3.2 Using the conda environment in the book’s repoA.3.3 Saving and loading environmentsA.4 Setting up your AWS EC2 environmentA.4.1 Creating an AWS accountA.4.2 Connecting remotely to your instanceA.4.3 Running your Jupyter notebook
index

Content preview from Deep Learning for Vision Systems

10 Visual embeddings

This chapter covers

Expressing similarity between images via loss functions
Training CNNs to achieve a desired embedding function with high accuracy
Using visual embeddings in real-world applications

by Ratnesh Kumar

Obtaining meaningful relationships between images is a vital building block for many applications that touch our lives every day, such as face recognition and image search algorithms. To tackle such problems, we need to build an algorithm that can extract relevant features from images and subsequently compare them using their corresponding features.

Ratnesh Kumar obtained his PhD from the STARS team at Inria, France, in 2014. While working on his PhD, he focused on problems in video understanding: video segmentation ...

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

Publisher Resources

ISBN: 9781617296192Publisher Support Publisher Website

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills

Deep Learning for Vision Systems

by Mohamed Elgendy

10 Visual embeddings

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.