Book Image

Hands-On Image Generation with TensorFlow

By : Soon Yau Cheong
Book Image

Hands-On Image Generation with TensorFlow

By: Soon Yau Cheong

Overview of this book

The emerging field of Generative Adversarial Networks (GANs) has made it possible to generate indistinguishable images from existing datasets. With this hands-on book, you’ll not only develop image generation skills but also gain a solid understanding of the underlying principles. Starting with an introduction to the fundamentals of image generation using TensorFlow, this book covers Variational Autoencoders (VAEs) and GANs. You’ll discover how to build models for different applications as you get to grips with performing face swaps using deepfakes, neural style transfer, image-to-image translation, turning simple images into photorealistic images, and much more. You’ll also understand how and why to construct state-of-the-art deep neural networks using advanced techniques such as spectral normalization and self-attention layer before working with advanced models for face generation and editing. You'll also be introduced to photo restoration, text-to-image synthesis, video retargeting, and neural rendering. Throughout the book, you’ll learn to implement models from scratch in TensorFlow 2.x, including PixelCNN, VAE, DCGAN, WGAN, pix2pix, CycleGAN, StyleGAN, GauGAN, and BigGAN. By the end of this book, you'll be well versed in TensorFlow and be able to implement image generative technologies confidently.
Table of Contents (15 chapters)
1
Section 1: Fundamentals of Image Generation with TensorFlow
5
Section 2: Applications of Deep Generative Models
9
Section 3: Advanced Deep Generative Techniques

Building a ProGAN

We have now learned about the three features of ProGANs – pixel normalization, minibatch standard deviation statistics, and the equalized learning rate. Now, we are going to delve into the network architecture and look at how to grow the network progressively. ProGAN grows an image by growing the layers, starting from a resolution of 4x4, then doubling it to 8x8, 16x16, and so on to 1024x1024. Thus, we will first write the code to build the layer block at each scale. The building blocks of the generator and discriminator are trivially simple, as we will see.

Building the generator blocks

We will start by building the 4x4 generator block, which forms the base of the generator and takes in the latent code as input. The input is normalized by PixelNorm before going to Dense. A lower gain is used for the equalized learning rate for that layer. Leaky ReLU and pixel normalization are used throughout all the generator blocks. We build the generator as follows...