Improved Precision and Recall Metric for Assessing Generative Models

TL;DR

This paper introduces a new evaluation metric for generative models that separately measures quality and coverage, demonstrating its effectiveness on StyleGAN and BigGAN, and analyzing model variants and truncation methods.

Contribution

The paper proposes a novel, explicit metric for assessing both quality and coverage of generative models, and provides insights into model architecture, training, and sampling techniques.

Findings

The new metric effectively distinguishes quality and coverage in image generation.

Identified improved StyleGAN variants with better sample properties.

Analyzed truncation methods, proposing an improved approach.

Abstract

The ability to automatically estimate the quality and coverage of the samples produced by a generative model is a vital requirement for driving algorithm research. We present an evaluation metric that can separately and reliably measure both of these aspects in image generation tasks by forming explicit, non-parametric representations of the manifolds of real and generated data. We demonstrate the effectiveness of our metric in StyleGAN and BigGAN by providing several illustrative examples where existing metrics yield uninformative or contradictory results. Furthermore, we analyze multiple design variants of StyleGAN to better understand the relationships between the model architecture, training methods, and the properties of the resulting sample distribution. In the process, we identify new variants that improve the state-of-the-art. We also perform the first principled analysis of truncation methods and identify an improved method. Finally, we extend our metric to estimate the perceptual quality of individual samples, and use this to study latent space interpolations.