schwettmann/visual-vocab

Pytorch-based tools for constructing a vocabulary of visual concepts in a GAN.

27
/ 100
Experimental

This tool helps researchers and vision scientists understand what visual concepts a Generative Adversarial Network (GAN) has learned. You input a pre-trained GAN and optionally your own image annotations, and it outputs a 'vocabulary' of human-interpretable visual concepts (like 'striped' or 'sky'). This is for anyone studying how AI models perceive and represent the visual world.

No commits in the last 6 months.

Use this if you want to extract and visualize the specific visual features or ideas a GAN has learned, expressed in plain language.

Not ideal if you're looking to train new image generation models from scratch or simply use a GAN for image synthesis without analyzing its internal representations.

AI interpretability computer vision research GAN analysis concept extraction machine learning explainability
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 6 / 25
Maturity 16 / 25
Community 5 / 25

How are scores calculated?

Stars

17

Forks

1

Language

Jupyter Notebook

License

MIT

Last pushed

Feb 25, 2022

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/schwettmann/visual-vocab"

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