Evaluating generative networks using Gaussian mixtures of image features

TL;DR

This paper introduces WaM, a new metric for evaluating generative models by modeling image features with Gaussian mixtures and computing the 2-Wasserstein distance, addressing limitations of the FID measure.

Contribution

The paper proposes WaM, a novel evaluation metric using GMMs and Wasserstein distance, improving accuracy over FID by capturing non-Gaussian feature distributions.

Findings

WaM is less sensitive to imperceptible image perturbations than FID.

Modeling features with GMMs provides a more accurate assessment of generative network performance.

WaM outperforms FID in various experimental settings.

Abstract

We develop a measure for evaluating the performance of generative networks given two sets of images. A popular performance measure currently used to do this is the Fr\'echet Inception Distance (FID). FID assumes that images featurized using the penultimate layer of Inception-v3 follow a Gaussian distribution, an assumption which cannot be violated if we wish to use FID as a metric. However, we show that Inception-v3 features of the ImageNet dataset are not Gaussian; in particular, every single marginal is not Gaussian. To remedy this problem, we model the featurized images using Gaussian mixture models (GMMs) and compute the 2-Wasserstein distance restricted to GMMs. We define a performance measure, which we call WaM, on two sets of images by using Inception-v3 (or another classifier) to featurize the images, estimate two GMMs, and use the restricted $2$-Wasserstein distance to compare the GMMs. We experimentally show the advantages of WaM over FID, including how FID is more sensitive than WaM to imperceptible image perturbations. By modelling the non-Gaussian features obtained from Inception-v3 as GMMs and using a GMM metric, we can more accurately evaluate generative network performance.