DGL-GAN: Discriminator Guided Learning for GAN Compression

TL;DR

This paper introduces DGL-GAN, a novel discriminator-guided learning method that effectively compresses large-scale GANs like StyleGAN2 and BigGAN, achieving state-of-the-art results while reducing computational costs.

Contribution

The paper proposes a simple discriminator-guided knowledge transfer approach for GAN compression, improving efficiency without sacrificing image quality.

Findings

DGL-GAN achieves SOTA results on StyleGAN2 and BigGAN.

DGL-GAN boosts the performance of uncompressed GANs, e.g., StyleGAN2 with FID 2.65.

The method effectively reduces computation costs while maintaining high-quality image synthesis.

Abstract

Generative Adversarial Networks (GANs) with high computation costs, e.g., BigGAN and StyleGAN2, have achieved remarkable results in synthesizing high-resolution images from random noise. Reducing the computation cost of GANs while keeping generating photo-realistic images is a challenging field. In this work, we propose a novel yet simple {\bf D}iscriminator {\bf G}uided {\bf L}earning approach for compressing vanilla {\bf GAN}, dubbed {\bf DGL-GAN}. Motivated by the phenomenon that the teacher discriminator may contain some meaningful information about both real images and fake images, we merely transfer the knowledge from the teacher discriminator via the adversarial interaction between the teacher discriminator and the student generator. We apply DGL-GAN to compress the two most representative large-scale vanilla GANs, i.e., StyleGAN2 and BigGAN. Experiments show that DGL-GAN achieves state-of-the-art (SOTA) results on both StyleGAN2 and BigGAN. Moreover, DGL-GAN is also effective in boosting the performance of original uncompressed GANs. Original uncompressed StyleGAN2 boosted with DGL-GAN achieves FID 2.65 on FFHQ, which achieves a new state-of-the-art performance. Code and models are available at \url{https://github.com/yuesongtian/DGL-GAN}