Probabilistic Precision and Recall Towards Reliable Evaluation of Generative Models

TL;DR

This paper critically analyzes existing kNN-based precision-recall metrics for generative models, identifies their limitations, and introduces probabilistic P-precision and P-recall metrics that offer more reliable evaluation of fidelity and diversity.

Contribution

The paper proposes novel probabilistic metrics, P-precision and P-recall, to improve the reliability of evaluating generative models over existing kNN-based methods.

Findings

PP&PR outperform existing metrics in reliability

Metrics are robust to outliers and distributional changes

Extensive experiments validate improved evaluation accuracy

Abstract

Assessing the fidelity and diversity of the generative model is a difficult but important issue for technological advancement. So, recent papers have introduced k-Nearest Neighbor ($k$NN) based precision-recall metrics to break down the statistical distance into fidelity and diversity. While they provide an intuitive method, we thoroughly analyze these metrics and identify oversimplified assumptions and undesirable properties of kNN that result in unreliable evaluation, such as susceptibility to outliers and insensitivity to distributional changes. Thus, we propose novel metrics, P-precision and P-recall (PP\&PR), based on a probabilistic approach that address the problems. Through extensive investigations on toy experiments and state-of-the-art generative models, we show that our PP\&PR provide more reliable estimates for comparing fidelity and diversity than the existing metrics. The codes are available at \url{https://github.com/kdst-team/Probablistic_precision_recall}.