---
title: "FID 隐藏了什么：生成式评估中偏差的检测、排序与诊断"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.24881v1"
permalink: "https://feed.astralor.com/items/itm_6d53abe717c83c77948115d7"
published: "2026-08-25T17:58:46.000Z"
score: 75
channel: "ai"
tags: ["DiT-XL/2","FID","SiT-XL/2","ZID","生成模型评估"]
---

# FID 隐藏了什么：生成式评估中偏差的检测、排序与诊断

## AI 点评

> 这篇论文值得所有认真做生成模型的团队读一下：FID 作为排行榜常客，其实可能被“欺骗”甚至给出与人类感知相反的结论。ZID 把单一标量拆成排序、检验和诊断三个输出，我觉得这才是评估指标该有的样子，尤其是高引导下多样性坍缩的欠离散标记，对调参很有参考价值。

## 速览

FID 和 KID 是常用的生成模型评估指标，但 FID 仅依赖前两阶矩，可能漏掉重要分布差异，且标量差距无法校准采样波动。本文提出 ZID（Z-resolved Integrated Diagnostic），结合六个标准化位置和离散度敏感指标，输出排序指数、置换检验 p 值和有符号离散度读数。实验显示 ZID 能检测 FID 失效的偏差，并在 DiT-XL/2 与 SiT-XL/2 引导扫描中识别高引导导致的多样性坍缩为欠离散。

## 关键点

- FID 仅匹配 Inception 均值和协方差，可被视觉上不可识别的图像欺骗：其 FID 24.7 优于真实图像 58.6。
- ZID 输出三个关联结果：偏离程度排序指数、分布相等性检验 p 值、有符号离散度诊断读数。
- 在受控实验中，ZID 能检测 FID 平坦或反向的多种偏差，且分数随严重程度单调变化。
- 在 DiT-XL/2 和 SiT-XL/2 引导扫描中，ZID 将高引导下的多样性坍缩诊断为欠离散（under-dispersion）。

## 全文

Generative models are commonly ranked by Fréchet Inception Distance (FID) and Kernel Inception Distance (KID), yet FID's first-two-moment summary can miss distributional differences, and a reported scalar gap alone is not a calibrated test against sampling variation. FID's moment restriction has concrete consequences: on ImageNet, visually unrecognizable images optimized only to match the reference Inception mean and covariance obtain FID $24.7$ versus $58.6$ for held-out real images (lower is better). Moreover, FID and KID are scalar discrepancies that are unchanged when the two samples are exchanged and therefore do not encode the direction of a dispersion change: under-dispersion, as can occur in mode collapse, versus over-dispersion. We introduce \textbf{ZID} (\emph{Z-resolved Integrated Diagnostic}), which combines six standardized location- and dispersion-sensitive arms from a rank graph (RISE) and Gaussian kernels (GPK at two bandwidths). Rather than asking one scalar to serve incompatible roles, ZID reports three linked outputs: an index for ranking departure magnitude, a permutation $p$-value for testing distributional equality, and a signed dispersion readout for diagnosis. In controlled experiments, ZID detects a broad range of departures, and its score tracks increasing severity along the corresponding sweeps, including cases in which FID is flat or reversed. On DiT-XL/2 and SiT-XL/2 guidance sweeps, ZID detects departure from real data, and its signed readout labels the high-guidance diversity collapse as under-dispersion.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_6d53abe717c83c77948115d7