marcojira/fld

PyTorch code for FLD (Feature Likelihood Divergence), FID, KID, Precision, Recall, etc. using DINOv2, InceptionV3, CLIP, etc.

31
/ 100
Emerging

This project helps machine learning engineers and researchers assess the quality of images generated by their generative AI models. It takes collections of real training images, real test images, and generated images as input. It then outputs various metrics, including Feature Likelihood Divergence (FLD), that tell you how good your generated images are in terms of fidelity, diversity, and crucially, novelty (i.e., whether the model is simply memorizing and copying its training data). This helps generative AI practitioners understand and improve their models.

No commits in the last 6 months.

Use this if you are developing or evaluating generative AI models, especially for images, and need to quantify how well they are generalizing and avoiding memorization.

Not ideal if you are not working with generative models or if you need to evaluate models on modalities other than images without appropriate feature extractors.

generative-ai image-synthesis model-evaluation machine-learning-research deep-learning
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 8 / 25
Maturity 8 / 25
Community 15 / 25

How are scores calculated?

Stars

63

Forks

10

Language

Python

License

Last pushed

Oct 08, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/marcojira/fld"

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