Assessing Generative Models via Precision and Recall

TL;DR

This paper introduces a new method to evaluate generative models by separately measuring their precision and recall, providing more detailed insights into their performance beyond existing single-score metrics.

Contribution

It proposes a novel, intuitive definition of precision and recall for distributions, along with an efficient algorithm for their computation, improving model evaluation.

Findings

The new metrics can distinguish between sample quality and coverage.

Empirical results show better disentanglement of quality and diversity.

The approach relates to existing metrics like FID and Inception Score.

Abstract

Recent advances in generative modeling have led to an increased interest in the study of statistical divergences as means of model comparison. Commonly used evaluation methods, such as the Frechet Inception Distance (FID), correlate well with the perceived quality of samples and are sensitive to mode dropping. However, these metrics are unable to distinguish between different failure cases since they only yield one-dimensional scores. We propose a novel definition of precision and recall for distributions which disentangles the divergence into two separate dimensions. The proposed notion is intuitive, retains desirable properties, and naturally leads to an efficient algorithm that can be used to evaluate generative models. We relate this notion to total variation as well as to recent evaluation metrics such as Inception Score and FID. To demonstrate the practical utility of the proposed approach we perform an empirical study on several variants of Generative Adversarial Networks and Variational Autoencoders. In an extensive set of experiments we show that the proposed metric is able to disentangle the quality of generated samples from the coverage of the target distribution.