Geometric GAN

Abstract

Generative Adversarial Nets (GANs) represent an important milestone for effective generative models, which has inspired numerous variants seemingly different from each other. One of the main contributions of this paper is to reveal a unified geometric structure in GAN and its variants. Specifically, we show that the adversarial generative model training can be decomposed into three geometric steps: separating hyperplane search, discriminator parameter update away from the separating hyperplane, and the generator update along the normal vector direction of the separating hyperplane. This geometric intuition reveals the limitations of the existing approaches and leads us to propose a new formulation called geometric GAN using SVM separating hyperplane that maximizes the margin. Our theoretical analysis shows that the geometric GAN converges to a Nash equilibrium between the discriminator and generator. In addition, extensive numerical results show that the superior performance of geometric GAN.

Cited by in corpus (34)

• Large Scale Adversarial Representation Learning
• Semantic Image Synthesis with Spatially-Adaptive Normalization
• Adversarial Video Generation on Complex Datasets
• SESAME: Semantic Editing of Scenes by Adding, Manipulating or Erasing Objects
• CopyCat: Many-to-Many Fine-Grained Prosody Transfer for Neural Text-to-Speech
• LOGAN: Latent Optimisation for Generative Adversarial Networks
• Generative Image Inpainting with Segmentation Confusion Adversarial Training and Contrastive Learning
• Composite Motion Learning with Task Control
• Improved Consistency Regularization for GANs
• Ultra-High-Resolution Detector Simulation with Intra-Event Aware GAN and Self-Supervised Relational Reasoning
• Learning Layout and Style Reconfigurable GANs for Controllable Image Synthesis
• Adma-GAN: Attribute-Driven Memory Augmented GANs for Text-to-Image Generation
• Triple consistency loss for pairing distributions in GAN-based face synthesis
• Anisotropic Stroke Control for Multiple Artists Style Transfer
• Transformation-based Adversarial Video Prediction on Large-Scale Data
• Text-only domain adaptation for end-to-end ASR using integrated text-to-mel-spectrogram generator
• cGANs with Multi-Hinge Loss
• Inception Score, Label Smoothing, Gradient Vanishing and -log(D(x)) Alternative
• DeshuffleGAN: A Self-Supervised GAN to Improve Structure Learning
• Understanding the Effectiveness of Lipschitz-Continuity in Generative Adversarial Nets
• Training Generative Reversible Networks
• Label-Noise Robust Generative Adversarial Networks
• Image Synthesis From Reconfigurable Layout and Style
• SocialInteractionGAN: Multi-person Interaction Sequence Generation
• CoNeS: Conditional neural fields with shift modulation for multi-sequence MRI translation
• Galaxy Imaging with Generative Models: Insights from a Two-Models Framework
• Model Imitation for Model-Based Reinforcement Learning
• Entropy-regularized Optimal Transport Generative Models
• A Novel Framework for Selection of GANs for an Application
• A Comprehensive Multi-scale Approach for Speech and Dynamics Synchrony in Talking Head Generation
• Perceiving Music Quality with GANs
• Noise Robust Generative Adversarial Networks
• Coulomb Autoencoders
• Realistic Face Reenactment via Self-Supervised Disentangling of Identity and Pose