mseitzer/pytorch-fid

Compute FID scores with PyTorch.

67
/ 100
Established

Ports the official TensorFlow FID implementation with PyTorch, computing Fréchet distance between Gaussian-fitted Inception network features to measure image dataset similarity. Supports flexible feature extraction from intermediate Inception layers (64–2048 dimensions) via global average pooling, enabling evaluation on smaller datasets, and can pre-compute and cache dataset statistics as `.npz` archives for efficient multi-model comparisons.

3,833 stars and 112,753 monthly downloads. No commits in the last 6 months. Available on PyPI.

Stale 6m
Maintenance 0 / 25
Adoption 20 / 25
Maturity 25 / 25
Community 22 / 25

How are scores calculated?

Stars

3,833

Forks

529

Language

Python

License

Apache-2.0

Last pushed

Jul 03, 2024

Monthly downloads

112,753

Commits (30d)

0

Dependencies

5

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/mseitzer/pytorch-fid"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.