Improved Consistency Regularization for GANs

TL;DR

This paper enhances consistency regularization techniques in GANs to improve image quality and stability, achieving state-of-the-art FID scores across multiple datasets and architectures.

Contribution

The authors identify artifacts caused by consistency regularization and propose modifications that significantly improve GAN performance and sample quality.

Findings

Achieved best known FID scores on CIFAR-10 and CelebA.

Improved conditional image synthesis FID from 11.48 to 9.21.

Enhanced BigGAN on ImageNet with FID from 6.66 to 5.38.

Abstract

Recent work has increased the performance of Generative Adversarial Networks (GANs) by enforcing a consistency cost on the discriminator. We improve on this technique in several ways. We first show that consistency regularization can introduce artifacts into the GAN samples and explain how to fix this issue. We then propose several modifications to the consistency regularization procedure designed to improve its performance. We carry out extensive experiments quantifying the benefit of our improvements. For unconditional image synthesis on CIFAR-10 and CelebA, our modifications yield the best known FID scores on various GAN architectures. For conditional image synthesis on CIFAR-10, we improve the state-of-the-art FID score from 11.48 to 9.21. Finally, on ImageNet-2012, we apply our technique to the original BigGAN model and improve the FID from 6.66 to 5.38, which is the best score at that model size.