Generative Adversarial Networks for Image Generation

Preface
Acknowledgments
Contents
Acronyms
1 Generative Adversarial Networks (GANs)
	1.1 Introduction to GANs
	1.2 Challenges of GANs
	Bibliography
2 GANs for Image Generation
	2.1 Image Generation
		2.1.1 Overview of Image Generation
		2.1.2 GANs for Image Generation
		2.1.3 Background Research of GANs
	2.2 Improving Quality for Generated Image with LSGANs
		2.2.1 Least Squares Generative Adversarial Networks
			2.2.1.1 Theoretical Analysis
			2.2.1.2 Parameter Selection
		2.2.2 Experiments
			2.2.2.1 Implementation Details
			2.2.2.2 Qualitative Evaluation
			2.2.2.3 Quantitative Evaluation
			2.2.2.4 Comparison of Two Parameter Schemes
	2.3 Improving Training Stability: Theoretical Analysis
		2.3.1 Approach
		2.3.2 Theoretical Analysis
		2.3.3 Experiments
			2.3.3.1 Evaluation Without Gradient Penalty
			2.3.3.2 Gaussian Mixture Distribution
			2.3.3.3 Difficult Architectures
			2.3.3.4 Datasets with Small Variability
			2.3.3.5 Evaluation with Gradient Penalty
		2.3.4 Discussion
	2.4 Multi-domain Image Generation with RCGANs
		2.4.1 Experiments
			2.4.1.1 Implementation Details
			2.4.1.2 Digits
			2.4.1.3 Edges and Photos
			2.4.1.4 Faces
			2.4.1.5 Quantitative Evaluation
			2.4.1.6 More Applications
	Bibliography
3 More Key Applications of GANs
	3.1 Image-to-Image Translation
		3.1.1 pix2pix
		3.1.2 CycleGAN
	3.2 Unsupervised Domain Adaptation
		3.2.1 Domain Adversarial Training
		3.2.2 Using Image-to-Image Translation
		3.2.3 Using RCGANs
			3.2.3.1 Experiments
	3.3 GANs for Security
	Bibliography
4 Conclusions
	4.1 Contributions
	4.2 Future Research
	Bibliography