Beyond FVD: Enhanced Evaluation Metrics for Video Generation Quality

TL;DR

This paper critically examines the limitations of the FVD metric for video quality evaluation and introduces JEDi, a new embedding distance that outperforms FVD in reliability, sample efficiency, and correlation with human judgment.

Contribution

The paper identifies key shortcomings of FVD and proposes JEDi, a novel metric based on Joint Embedding Predictive Architecture, demonstrating superior performance in multiple datasets.

Findings

JEDi requires only 16% of samples compared to FVD.

JEDi increases alignment with human evaluation by 34%.

FVD's limitations include non-Gaussian features and insensitivity to temporal distortions.

Abstract

The Fr\'echet Video Distance (FVD) is a widely adopted metric for evaluating video generation distribution quality. However, its effectiveness relies on critical assumptions. Our analysis reveals three significant limitations: (1) the non-Gaussianity of the Inflated 3D Convnet (I3D) feature space; (2) the insensitivity of I3D features to temporal distortions; (3) the impractical sample sizes required for reliable estimation. These findings undermine FVD's reliability and show that FVD falls short as a standalone metric for video generation evaluation. After extensive analysis of a wide range of metrics and backbone architectures, we propose JEDi, the JEPA Embedding Distance, based on features derived from a Joint Embedding Predictive Architecture, measured using Maximum Mean Discrepancy with polynomial kernel. Our experiments on multiple open-source datasets show clear evidence that it is a superior alternative to the widely used FVD metric, requiring only 16% of the samples to reach its steady value, while increasing alignment with human evaluation by 34%, on average.