ViTGAN: Training GANs with Vision Transformers
This addresses the challenge of applying ViTs to generative tasks for the computer vision community, representing an incremental advancement by adapting existing GAN frameworks with new regularization techniques.
The paper tackled the problem of extending Vision Transformers (ViTs) to image generation by integrating them into GANs, achieving comparable performance to leading CNN-based GAN models on datasets like CIFAR-10, CelebA, and LSUN bedroom.
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.