Discriminator Rejection Sampling

TL;DR

This paper introduces Discriminator Rejection Sampling (DRS), a practical algorithm that leverages GAN discriminators to improve sample quality and approximate the true data distribution more closely.

Contribution

The paper presents a novel rejection sampling method using GAN discriminators, with theoretical analysis and practical implementation that enhances image generation quality.

Findings

DRS improves Inception Score on ImageNet from 52.52 to 76.08.

DRS reduces FID on ImageNet from 18.65 to 13.75.

Baseline model improvements increase Inception Score from 52.52 to 62.36.

Abstract

We propose a rejection sampling scheme using the discriminator of a GAN to approximately correct errors in the GAN generator distribution. We show that under quite strict assumptions, this will allow us to recover the data distribution exactly. We then examine where those strict assumptions break down and design a practical algorithm - called Discriminator Rejection Sampling (DRS) - that can be used on real data-sets. Finally, we demonstrate the efficacy of DRS on a mixture of Gaussians and on the SAGAN model, state-of-the-art in the image generation task at the time of developing this work. On ImageNet, we train an improved baseline that increases the Inception Score from 52.52 to 62.36 and reduces the Frechet Inception Distance from 18.65 to 14.79. We then use DRS to further improve on this baseline, improving the Inception Score to 76.08 and the FID to 13.75.