Records · kv:2610.00018v1

Reproduction of atong01/conditional-flow-matching: OT-CFM CIFAR-10 FID

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-08

Independent reproduction · Artifact evaluation · Computer vision

PaperReproduction of: original code

Abstract

Evaluated the authors' released OT-CFM CIFAR-10 weights (400,000 training steps, release 1.0.4) with the repository's compute_fid.py and the README settings at the pinned commit: 50,000 samples generated with the adaptive dopri5 solver (tolerance 1e-5) and FID computed against the CIFAR-10 training set with clean-fid. Training was not repeated (400,000 steps exceeds the time budget); the weights file is attested by hash in the report. SciPy was pinned below 1.16 for clean-fid compatibility. One NVIDIA RTX 5060 Ti.

Sealed report

Cite this record

@misc{kveritas261000018v1,
  title={Reproduction of atong01/conditional-flow-matching: OT-CFM CIFAR-10 FID},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00018v1},
  url={https://kveritas.org/records/2610.00018v1},
}

References

1. [1] Mamadou K. Keita and Christopher Homan. Computer Science Conferences Should Require Nonrepudiable Experimental Results. NeurIPS 2026 Position Paper Track. arXiv:2605.08586.

Sealed with K-Veritas [1]. Listed, not endorsed. A record shows these results came from the sealed code, unchanged. Whether the work is sound is for the reader to judge.