Generative Adversarial Networks with Industrial Use Cases
US$ 19.95
The publisher has enabled DRM protection, which means that you need to use the BookFusion iOS, Android or Web app to read this eBook. This eBook cannot be used outside of the BookFusion platform.
Description
Contents
Reviews
Language
English
ISBN
9789389423853
Cover Page
Title Page
Copyright Page
Dedication
About the Author
Acknowledgement
Preface
Errata
Table of Contents
1. Basics of Generative Adversarial Networks (GAN)
Introduction
Structure
Objectives
Deep learning applications at a glance
Types of deep learning applications
Image classification
Semantic segmentation
Semantic search
Text classification
Generating images
Generator
Discriminator
Object detection
Multi-variate time series prediction
Information extraction from the scan
Different deep learning frameworks
Introduction to GAN
GAN architecture and explanation
Getting started with GAN
Task
Dataset
Data extraction
Preprocessing the images
Architecture of generator
Architecture of discriminator
Noise
Loss
How does it work?
Optimizer
Training
Output
Conclusion
2. GAN Applications
Introduction
Structure
Objective
Health sectorspecific GAN applications
Health data generation
Generating multi-label discrete patient records using GAN
MedGAN
Medical image translation using GANs
MedGANarchitecture
pix2pix
PAN
ID-CGAN
ID-CGAN architecture
Fila-SGAN
Synthetic medical images from dual generative adversarial networks
Why do we need?
seGAN - medical image segmentation
Anomaly detection
Retail sector-specific GAN applications
SRGAN
Virtual try-on clothes
Denoising images
Sketch to (colorful/realistic)handbag/shoes
Pose guided person image generation
PixelDTGAN-taking clothes from celebrity pictures
Disco GAN
pix2pix - removing the filter from face
Media and entertainment sector-specific GAN applications
SRGAN
DeOldify
Stack GAN
Face ageing - Age-cGAN
Domain Transfer Networks (DTN)
DTN architecture
Autonomous vehicles sector-specific GAN applications
Autonomous driving testing system
EduTechsector-specific GAN applications
pix2pix
SRGAN
Denoising GAN
Image editing - IcGAN
Telecom sector-specific GAN applications
WaveNet
GANSynth
Mixed reality specific GAN applications
Conclusion
References
3. Problem with GAN
Introduction
The objective function for training GAN
KL divergence
KL divergence between two discrete probability distributions
KL Divergence between two continuous probability distributions
Nash equilibrium
Prisoner’s dilemma
Example
The mentality behind solving GAN training problems
Mode collapse
Instability of adversarial training
Lack of a proper evaluation metric
Vanishing gradient
Improving training
One-sided label smoothing
Virtual batch normalization (VBN)
Adding noises
Minibatch discrimination
Use a better metric of distribution similarity
Wasserstein GAN (WGAN)
Wasserstein distance
Why is Wasserstein better than JS or KL divergence?
Wasserstein distance as GAN loss function
Conclusion
4. Famous Types of GANs
Structure
Objective
Generative Adversarial Network
Training GAN’s
Conditional Generative Adversarial Network (cGAN)
Applications
Architecture
Mini-Maxcondition in cGAN
LOSS
DCGAN
The generator
The discriminator
Applications
InfoGAN
Coding in InfoGAN
Theory
Architecture
Results
Pix2Pix
Components of Pix2Pix model
Architecture
Generator
Discriminator
Results
pix2pix application
PAN
ID-CGAN
Stack GAN
Conditioning augmentation
Stage I
Stage II
Cycle GAN
CycleGAN transfers styles to images
Coding implementation:
Style GAN
Mapping network
Style modules (AdaIN)
Removing initial input
Stochastic variation
Style mixing
Truncation trick in W
Radial GAN
Conclusion
Exercise
The book hasn't received reviews yet.
