ViTGAN: Training GANs with Vision Transformers
arXiv:2107.04589
Abstract
Recently, Vision Transformers (ViTs) have shown competitive performance on image recognition while requiring less vision-specific inductive biases. In this paper, we investigate if such performance can be extended to image generation. To this end, we integrate the ViT architecture into generative adversarial networks (GANs). For ViT discriminators, we observe that existing regularization methods for GANs interact poorly with self-attention, causing serious instability during training. To resolve this issue, we introduce several novel regularization techniques for training GANs with ViTs. For ViT generators, we examine architectural choices for latent and pixel mapping layers to facilitate convergence. Empirically, our approach, named ViTGAN, achieves comparable performance to the leading CNN-based GAN models on three datasets: CIFAR-10, CelebA, and LSUN bedroom.
Accepted to ICLR 2022 (Spotlight)
References in corpus (19)
- GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
- Large Scale GAN Training for High Fidelity Natural Image Synthesis
- Spectral Normalization for Generative Adversarial Networks
- Self-Attention Generative Adversarial Networks
- Is Space-Time Attention All You Need for Video Understanding?
- MLP-Mixer: An all-MLP Architecture for Vision
- LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop
- Zero-Shot Text-to-Image Generation
- Which Training Methods for GANs do actually Converge?
- TransGAN: Two Pure Transformers Can Make One Strong GAN, and That Can Scale Up
- On Convergence and Stability of GANs
- Toward Transformer-Based Object Detection
- Pre-Trained Image Processing Transformer
- Image Augmentations for GAN Training
- Train Sparsely, Generate Densely: Memory-efficient Unsupervised Training of High-resolution Temporal GAN
- Multi-Head Attention: Collaborate Instead of Concatenate
- On Self Modulation for Generative Adversarial Networks
- Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth
- Lipschitz Generative Adversarial Nets