titleThe FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation
authorsNicolas Dufour, Alexei A. Efros, Patrick Pérez
date2026-06-18
venuearXiv (preprint, not peer reviewed)
idarXiv:2606.20536
affiliationKyutai
upvotes10 (Hugging Face Daily Papers, 2026-11-01)
linksabs pdf hf
abstractThe Frechet Inception Distance (FID) is the de facto arbiter of image generation, yet most papers report just a single number from a single trained model using a single sampling seed. How reproducible is that number if we retrain the model, or merely resample from it? In this paper, we treat FID as a random variable on a two-axis panel of training and generation seeds, and measure its variance directly on several hundred SiT networks trained on class-conditional ImageNet 256x256. We report surprising findings: (a) Retraining the model using the same recipe with a different seed moves FID 3.2x more (in Inception feature space) than redrawing samples from a fixed network. (b) That gap is driven by three factors: random initialisation, data ordering, and the per-step Gaussian noise of the flow-matching loss. (c) Increasing compute or model size barely tightens the spread, holding the FID coefficient of variation (CoV) inside a 1-2% band. (d) Per-cell classifier-free-guidance tuning halves the spread but reshuffles which seeds work best, and a lucky training seed reaches the same FID with up to 2x less compute than an unlucky one. Based on these findings, we recommend a new FID evaluation protocol: evaluate under per-cell optimal guidance, treat any FID gap below the empirically measured ~1.3% CoV as inconclusive, and report an error bar over several training seeds rather than a single FID number.

cite

@misc{dufour2026fid,
  title={The FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation},
  author={Nicolas Dufour and Alexei A. Efros and Patrick Pérez},
  year={2026},
  eprint={2606.20536},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2606.20536}
}