mseitzer/pytorch-fid

Health v2 · maintenance only

Adoption not part of the score

3851 stars · 528 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

A PyTorch port of the official TensorFlow implementation of the Fréchet Inception Distance (FID), a metric for measuring similarity between two image datasets. It is primarily used to evaluate the visual quality of samples from generative models like GANs.

Use cases

• compute FID score between two image folders
• evaluate GAN sample quality with PyTorch
• compare generated images to real dataset distribution
• calculate FID on GPU
• use different Inception feature layers for small datasets
• reproduce FID scores comparable to the official implementation

When to choose

• you train GANs or diffusion models in PyTorch and need FID evaluation
• you want a pip-installable, scriptable FID calculator
• your datasets are small and you need alternative Inception feature layers
• you need results closely matching the official TensorFlow FID implementation

When to avoid

• you need FID scores exactly comparable to papers using the official TensorFlow implementation
• you need other generative evaluation metrics like Inception Score or KID
• you don't use PyTorch and prefer the original TensorFlow tooling

Facets

library · maturity stable

machine-learning image-processing benchmarking cli deep-learning machine-learning computer-vision image-processing python cross-platform pytorch fid-score frechet-inception-distance gan-evaluation inception-network generative-models gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="mseitzer/pytorch-fid")