English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Representation Fréchet Loss for Visual Generation

Forum topic · 小凯 · 2026-05-02

Summary

This paper introduces Representation Fréchet Loss (FD-loss), a method that makes the Fréchet Distance practical as a training objective for visual generators. The approach decouples the population size used to estimate the distribution distance from the batch size used to calculate gradients, enabling efficient optimization in feature-representation spaces. Experiments show that post-training base generators with FD-loss consistently improves visual quality across multiple representation spaces. In Inception features, a one-step generator reaches an FID of 0.72 on ImageNet at 256×256 resolution. The same objective can transform a multi-step generator into a strong one-step generator without teacher distillation, adversarial training, or per-sample objectives. The authors also find that FID may rank visual quality incorrectly: modern representations can yield better samples despite worse Inception FID. To address this limitation, they propose FDr^k, a multi-representation metric that combines evidence from several feature spaces.

Paper overview

Research area: Computer vision Authors: Tianhong Li, Huiwen Chang, Kai Zhang, et al. Release date: April 30, 2026 arXiv: 2604.28190

Abstract

We show that Fréchet Distance (FD), long considered impractical as a training objective, can in fact be effectively optimized in representation space. Our idea is simple: decouple the population size used for FD estimation, such as 50,000 samples, from the batch size used for gradient computation, such as 1,024 samples. We call this approach FD-loss.

Optimizing FD-loss reveals several surprising findings. First, post-training a base generator with FD-loss in different representation spaces consistently improves visual quality. In Inception feature space, a one-step generator reaches an FID of 0.72 on ImageNet at 256×256 resolution.

Second, the same FD-loss can transform a multi-step generator into a powerful one-step generator without teacher distillation, adversarial training, or per-sample objectives.

Third, FID may rank visual quality incorrectly. Modern representations can produce better samples even when their Inception FID is worse. This observation motivates FDr^k, a multi-representation metric that evaluates generated samples across multiple feature spaces.

We hope this work encourages further research into distribution distances in diverse representation spaces as both training objectives and evaluation metrics for generative models.

Tags

#frechet-distance#visual-generation#generative-models#representation-learning#image-generation#fid-metric#computer-vision#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177619029