This book explores how to use Generative Adversarial Network (GANs) in a variety of applications and emphasises their substantial advancements over traditional generative models. This book's major goal is to concentrate on cutting-edge research in deep learning and generative adversarial networks, which includes creating new tools and methods for processing text, images, and audio.
A Generative Adversarial Network (GAN) is a class of machine learning framework and is the next emerging network in deep learning applications. Generative Adversarial Networks(GANs) have the feasibility to build improved models, as they can generate the sample data as per application requirements. There are various applications of GAN in science and technology, including computer vision, security, multimedia and advertisements, image generation, image translation, text-to-images synthesis, video synthesis, generating high-resolution images, drug discovery, etc.
A convolutional neural network or a recurrent neural network can be used as the discriminator network, while a de-convolutional neural network can be used as the generator network. As a result, GANs can be used to build multidimensional data distributions like pictures. GANs have shown potential in a variety of difficult generative tasks, including text-to-photo translation, picture generation, image composition, and image-to-image translation. GANs are a powerful type of deep generative model; however, they have a variety of training issues, such as mode collapse and training instability. There are different types of learning approaches in machine learning such as supervised and unsupervised learning.
Unsupervised Learning is a technique for teaching computers to use data that has not been classified or labeled. It means that no preparation information is accessible and the machine is customized to learn all alone. With no earlier information on the information, the machine should have the option to group it. The objective is to open the machines to huge measures of different information and allow them to gain from it to uncover already obscure bits of knowledge and reveal stowed away examples. Accordingly, solo learning calculations don’t necessarily create unsurprising outcomes. Rather, it figures out what makes the given dataset novel or interesting. It is important to program the machine to learn all alone. Both organized and unstructured information should be perceived and examined by the PC. Solo learning calculations can deal with more complicated handling undertakings than regulated learning frameworks.
Features:
Presents a comprehensive guide on how to use GAN for images and videos.
Includes case studies of Underwater Image Enhancement Using Generative Adversarial Network, Intrusion detection using GAN
Highlights the inclusion of gaming effects using deep learning methods
Examines the significant technological advancements in GAN and its real-world application.
Discusses as GAN challenges and optimal solutions
The book addresses scientific aspects for a wider audience such as junior and senior engineering, undergraduate and postgraduate students, researchers, and anyone interested in the trends development and opportunities in GAN and Deep Learning.
The material in the book can serve as a reference in libraries, accreditation agencies, government agencies, and especially the academic institution of higher education intending to launch or reform their engineering curriculum.
Author(s): Roshani Raut, Pranav D Pathak, Sachin R Sakhare, Sonali Patil
Publisher: CRC Press
Year: 2023
Language: English
Commentary: true
Pages: 223
Cover
Half Title
Title Page
Copyright Page
Table of Contents
Preface
Editors
List of Contributors
1 Generative Adversarial Networks and Its Use Cases
1.1 Introduction
1.2 Supervised Learning
1.2.1 Unsupervised Learning
1.3 Background of GAN
1.3.1 Image-To-Image Translation
1.4 Difference Between Auto Encoders and Generative Adversarial Networks
1.4.1 Auto Encoders
1.4.2 Generative Adversarial Networks
1.5 Difference Between VAN and Generative Adversarial Networks
1.6 Application of GANs
1.6.1 Application of GANs in Healthcare
1.6.2 Applications of Generative Models
1.6.2.1 Generate Examples for Image Datasets
1.6.2.2 Generate Realistic Photographs
1.6.2.3 Generate Cartoon Characters
1.6.2.4 Image-To-Image Translation
1.6.2.5 Text-To-Image Translation
1.6.2.6 Semantic-Image-To-Photo Translation
1.6.2.7 Photos to Emojis
1.6.2.8 Photograph Editing
1.6.2.9 Face Aging
1.7 Conclusion
References
2 Image-To-Image Translation Using Generative Adversarial Networks
2.1 Introduction
2.2 Conventional I2I Translations
2.2.1 Filtering-Based I2I
2.2.2 Optimisation-Based I2I
2.2.3 Dictionary Learning-Based I2I
2.2.4 Deep Learning-Based I2I
2.2.5 GAN-Based I2I
2.3 Generative Adversarial Networks (GAN)
2.3.1 How GANs Work
2.3.2 Loss Functions
2.3.2.1 Minimax Loss
2.3.3 Other Generative Models
2.4 Supervised I2I Translation
2.4.1 Pix2Pix
2.4.1.1 Applications of Pix2Pix Models
2.4.2 Additional Work On Supervised I2I Translations
2.4.2.1 Single-Modal Outputs
2.4.2.2 Multimodal Outputs
2.5 Unsupervised I2I (UI2I) Translation
2.5.1 Deep Convolutional GAN (DCGAN)
2.5.1.1 DCGAN Applications
2.5.2 Conditional GAN (CGAN)
2.5.3 Cycle GAN
2.5.3.1 Cycle Consistency Loss
2.5.3.2 CycleGAN Applications
2.5.4 Additional Work On Unsupervised I2I
2.5.4.1 Single-Modal Outputs
2.6 Semi-Supervised I2I
2.7 Few-Shot I2I
2.8 Comparative Analysis
2.8.1 Metrics
2.8.2 Results
2.9 Conclusion
References
3 Image Editing Using Generative Adversarial Network
3.1 Introduction
3.2 Background of GAN
3.3 Image-To-Image Translation
3.4 Motivation and Contribution
3.5 GAN Objective Functions
3.5.1 GAN Loss Challenges
3.5.2 The Problem of GAN Loss
3.5.3 Loss of Discriminator
3.5.4 GAN Loss Minimax
3.6 Image-To-Image Translation
3.6.1 Controlled Image-To-Image Conversion
3.6.1.1 CGAN
3.6.1.2 BicycleGAN
3.6.1.3 SPA-GAN
3.6.1.4 CE-GAN
3.6.2 Unsupervised Image to Image Conversion
3.6.2.1 CycleGAN
3.6.2.2 Dugan
3.6.2.3 UNIT
3.6.2.4 MUNIT
3.7 Application
3.8 Conclusion
References
4 Generative Adversarial Networks for Video-To-Video Translation
4.1 Introduction
4.2 Description of Background
4.2.1 Objectives
4.3 Different Methods and Architectures
4.4 Architecture
4.4.1 Cycle GAN
4.4.2 Style GAN
4.4.3 LS-GAN
4.4.4 Disco GAN
4.4.5 Mo-Cycle GAN
4.4.6 Different GANs for Video Synthesis (Fixed Length)
4.4.7 TGAN
4.4.8 Generative Adversarial Network: Flexible Dimension Audiovisual Combination
4.4.8.1 MoCo GAN
4.4.8.2 DVD GAN
4.4.8.3 Methods and Tools for GAN
4.4.8.4 GAN Lab
4.4.9 Hyper GAN
4.4.10 Imaginaire
4.4.11 GAN Tool Compartment
4.4.12 Mimicry
4.4.13 Pygan
4.4.14 Studio GAN
4.4.15 Torch GAN
4.4.16 TF-GAN
4.4.17 Ve GANs
4.5 Conclusions
References
5 Security Issues in Generative Adversarial Networks
5.1 Introduction
5.2 Motivation
5.2.1 Objectives
5.3 Related Work
5.3.1 Generative Adversarial Network
5.3.2 Overview of Security
5.3.3 GANs in Safety
5.3.3.1 Obscuring Delicate Information
5.3.4 Cyber Interruption and Malware Detection
5.3.5 Security Examination
5.4 Security Attacks in GANs
5.4.1 Cracking Passphrases
5.4.2 Hiding Malware
5.4.3 Forging Facial Detection
5.4.4 Detection and Response
5.5 Conclusion
References
6 Generative Adversarial Networks-Aided Intrusion Detection System
6.1 Introduction
6.2 Application of GANs for Resolving Data Imbalance
6.3 Application of GAN as a Deep Learning Classifier
6.4 Application of GANs for Generating Adversarial Examples
6.5 Conclusion
Glossary of Terms, Acronyms and Abbreviations
References
7 Textual Description to Facial Image Generation
7.1 Introduction
7.2 Literature Review
7.3 Dataset Description
7.4 Proposed Methodology
7.4.1 Generator
7.4.1.1 DAN (Deep Averaging Network)
7.4.1.2 Transformer Encoder
7.4.2 Discriminator
7.4.3 Training of GAN
7.4.3.1 Loss Function
7.4.3.2 Optimizer
7.4.3.3 Discriminative Learning Rates
7.4.3.4 Dropout
7.5 Limitations
7.6 Future Scope
7.7 Conclusion
7.8 Applications
References
8 An Application of Generative Adversarial Network in Natural Language Generation
8.1 Introduction
8.2 Generative Adversarial Network Model
8.2.1 Working of Generative Adversarial Network
8.2.2 Natural Language Generation
8.3 Background and Motivation
8.4 Related Work
8.5 Issues and Challenges
8.6 Case Studies: Application of Generative Adversarial Network
8.6.1 Creating Machines to Paint, Write, Compose, and Play
8.6.2 Use of GAN in Text Generation
8.6.3 Indian Sign Language Generation Using Sentence Processing and Generative Adversarial Networks
8.6.4 Applications of GAN in Natural Language Processing
8.7 Conclusions
References
9 Beyond Image Synthesis: GAN and Audio
9.1 Introduction
9.1.1 Audio Signals
9.2 About GANs
9.3 Working Principal of GANs
9.4 Literatutre Survey About Different GANs
9.4.1 Time Sequence Gan Adversarial Network
9.4.2 Vector-Quantized Contrastive Predictive Coding-GAN
9.4.3 The VQCPC Encoder
9.4.4 The Generative Adversarial Network Designs
9.5 Results
9.5.1 Dataset
9.5.2 Assessment
9.6 Baselines
9.7 Quantitative Outcomes
9.8 Casual Tuning In
9.9 Results
References
10 A Study On the Application Domains of Electroencephalogram for the Deep Learning-Based Transformative Healthcare
10.1 Introduction
10.2 Modalities of Deep Learning-Based Healthcare Applications
10.2.1 Medical Image Generation and Synthesis
10.2.2 EEG Signal Reconstruction and SSVEP Classification
10.2.3 Body Sensor-Induced Healthcare Applications
10.3 Healthcare Application Areas of EEG
10.3.1 Rare Disease Diagnosis
10.3.2 Robotics-Based Applications of Deep Learning Inducing EEG
10.3.3 Rehabilitation
10.3.3.1 Bipolar Disorder
10.3.3.2 Drug Rehabilitation
10.3.3.3 Gait Rehabilitation
10.3.3.4 Vascular Hemiplegia Rehabilitation
10.3.3.5 Dementia
10.3.3.6 Epilepsy
10.4 Significance of Different Electrode Placement Techniques
10.4.1 10–20 International System
10.4.2 10–10 System
10.4.3 10–5 System
10.5 Conclusion
References
11 Emotion Detection Using Generative Adversarial Network
11.1 Introduction
11.2 Background Study
11.3 Deep Learning Methods Used in Gaming Applications
11.3.1 Super-Resolution GAN
11.3.2 Deep Convolutional Generative Adversarial Network (DC-GAN)
11.3.3 Conditional Embedding Self-Attention Generative Adversarial Network
11.3.4 Variational Autoencoders Generative Adversarial Network
11.3.5 Conditional Generative Adversarial Network (CGAN)
11.3.6 Token-Based One-Shot Arbitrary Dimension Generative Adversarial Network
11.3.7 Poker-Face Generative Adversarial Network
11.4 Application Areas
11.4.1 Quality Enhancement
11.4.2 Differential Rendering
11.4.3 Character Auto-Creation and Customization
11.4.4 Procedural Content Generation
11.4.5 Video Game Evaluation
11.4.6 User Emotion Identification
11.5 Conclusion
References
12 Underwater Image Enhancement Using Generative Adversarial Network
12.1 Introduction
12.2 Literature Review
12.3 Proposed Method
12.3.1 Loss Function
12.3.2 Discriminator Loss
12.3.3 Generator Loss
12.4 Generative Adversarial Networks
12.5 Atrous Convolution
12.6 Experimental Results
12.6.1 Underwater Image Quality Measure (UIQM)
12.7 Conclusions
References
13 Towards GAN Challenges and Its Optimal Solutions
13.1 Introduction: Background and Driving Forces
13.2 Challenges With GAN
13.3 GAN Training Problems
13.3.1 NashEquilibrium
13.3.2 Vanishing Gradient
13.3.2.1 Mode Collapse and Non-Convergence
13.4 Conclusion
References
Index
