Hands-On Generative Adversarial Networks with PyTorch 1.x

Cover
Title Page
Copyright and Credits
About Packt
Contributors
Table of Contents
Preface
Section 1: Introduction to GANs and PyTorch
Chapter 1: Generative Adversarial Networks Fundamentals
Fundamentals of machine learning
Machine learning – classification and generation
Introducing adversarial learning
Generator and discriminator networks
Mathematical background of GANs
Using NumPy to train a sine signal generator
Designing the network architectures
Defining activation functions and the loss function
Working on forward pass and backpropagation
Training our GAN model
What GAN we do?
Image processing
Image synthesis
Image translation
Video synthesis and translation
NLP
3D modeling
Summary
References and useful reading list
Chapter 2: Getting Started with PyTorch 1.3
What's new in PyTorch 1.3?
Easy switching from eager mode to graph mode
The C++ frontend
The redesigned distributed library
Better research reproducibility
Miscellaneous
The PyTorch ecosystem
Cloud support
Migrating your previous code to 1.x
CUDA – GPU acceleration for fast training and evaluation
Installing NVIDIA driver
Installing CUDA
Installing cuDNN
Evaluating your CUDA installation
Installing PyTorch on Windows and Linux
Setting up the Python environment
Installing Python
Installing Anaconda Python
Prerequisites before we move on
Installing PyTorch
Installing official binaries
Building Pytorch from source
Evaluating your PyTorch installation
Bonus: setting up VS Code for Python coding
Configuring VS Code for Python development
Recommended VS Code extensions
References and useful reading list
Summary
Chapter 3: Best Practices for Model Design and Training
Model design cheat sheet
Overall model architecture design
Choosing a convolution operation method
Choosing a downsampling operation method
More on model design
Model training cheat sheet
Parameter initialization
Adjusting the loss function
Choosing an optimization method
Adjusting the learning rate
Gradient clipping, weight clipping, and more
Efficient coding in Python
Reinventing the wheel wisely
Advice for beginners in deep learning
Summary 
Section 2: Typical GAN Models for Image Synthesis
Chapter 4: Building Your First GAN with PyTorch
Introduction to Deep Convolutional GANs
The architecture of generator
The architecture of a discriminator
Creating a DCGAN with PyTorch
Generator network
Discriminator network
Model training and evaluation
Training iteration
Visualizing generated samples
Checking GPU usage information
Moving to larger datasets
Generating human faces from the CelebA dataset
Generating bedroom photos from the LSUN dataset
Having fun with the generator network
Image interpolation
Semantic vector arithmetic
Summary
References and useful reading list
Chapter 5: Generating Images Based on Label Information
CGANs – how are labels used?
Combining labels with the generator
Integrating labels into the discriminator
Generating images from labels with the CGAN
One-stop model training API
Argument parsing and model training
Working with Fashion-MNIST
InfoGAN – unsupervised attribute extraction
Network definitions of InfoGAN
Training and evaluation of InfoGAN
References and useful reading list
Summary
Chapter 6: Image-to-Image Translation and Its Applications
Using pixel-wise labels to translate images with pix2pix
Generator architecture
Discriminator architecture
Training and evaluation of pix2pix
Pix2pixHD – high-resolution image translation
Model architecture
Model training
CycleGAN – image-to-image translation from unpaired collections
Cycle consistency-based model design
Model training and evaluation
Summary
Furthering reading
Chapter 7: Image Restoration with GANs
Image super-resolution with SRGAN
Creating a generator
Creating the discriminator
Defining training loss
Training SRGAN to generate high-resolution images
Generative image inpainting
Efficient convolution – from im2col to nn.Unfold
WGAN – understanding the Wasserstein distance
Analyzing the problems with vanilla GAN loss
The advantages of Wasserstein distance
Training GAN for image inpainting
Model design for image inpainting
Implementation of Wasserstein loss
Summary
Useful reading list and references
Chapter 8: Training Your GANs to Break Different Models
Adversarial examples – attacking deep learning models
What are adversarial examples and how are they created?
Adversarial attacking with PyTorch
Generative adversarial examples
Preparing an ensemble classifier for Kaggle's Cats vs. Dogs
Breaking the classifier with advGAN
Summary
References and further reading list
Chapter 9: Image Generation from Description Text
Text-to-image synthesis with GANs
Quick introduction to word embedding
Translating text to image with zero-shot transfer learning
Zero-shot learning
GAN architecture and training
Generating photo-realistic images with StackGAN++
High-resolution text-to-image synthesis with StackGAN
From StackGAN to StackGAN++
Training StackGAN++ to generate images with better quality
Summary 
Further reading
Chapter 10: Sequence Synthesis with GANs
Text generation via SeqGAN – teaching GANs how to tell jokes
Design of SeqGAN – GAN, LSTM, and RL
A quick introduction to RNN and LSTM
Reinforcement learning versus supervised learning
Architecture of SeqGAN
Creating your own vocabulary for training
Speech quality enhancement with SEGAN
SEGAN architecture
Training SEGAN to enhance speech quality
Summary
Further reading
Chapter 11: Reconstructing 3D models with GANs
Fundamental concepts in computer graphics
Representation of 3D objects
Attributes of a 3D object
Camera and projection
Designing GANs for 3D data synthesis
Generators and discriminators in 3D-GAN
Training 3D-GAN
Summary
Further reading
Other Books You May Enjoy
Index