Deceive D: Adaptive Pseudo Augmentation for GAN Training with Limited Data

TL;DR

This paper proposes Adaptive Pseudo Augmentation (APA), a novel method that uses generated images to augment real data, improving GAN training with limited data by reducing discriminator overfitting and enhancing image synthesis quality.

Contribution

APA introduces a new adaptive augmentation strategy that employs the generator to augment real data, addressing overfitting in low-data GAN training and seamlessly integrating with existing models.

Findings

APA improves image quality in low-data regimes

Theoretical analysis supports convergence of the method

APA adds negligible computational overhead

Abstract

Generative adversarial networks (GANs) typically require ample data for training in order to synthesize high-fidelity images. Recent studies have shown that training GANs with limited data remains formidable due to discriminator overfitting, the underlying cause that impedes the generator's convergence. This paper introduces a novel strategy called Adaptive Pseudo Augmentation (APA) to encourage healthy competition between the generator and the discriminator. As an alternative method to existing approaches that rely on standard data augmentations or model regularization, APA alleviates overfitting by employing the generator itself to augment the real data distribution with generated images, which deceives the discriminator adaptively. Extensive experiments demonstrate the effectiveness of APA in improving synthesis quality in the low-data regime. We provide a theoretical analysis to examine the convergence and rationality of our new training strategy. APA is simple and effective. It can be added seamlessly to powerful contemporary GANs, such as StyleGAN2, with negligible computational cost.