KempnerInstitute/overcomplete

👋 Overcomplete is a Vision-based SAE Toolbox

39
/ 100
Emerging

This library helps machine learning researchers and practitioners understand what visual concepts large AI models are learning. It takes the internal activations from a vision model and extracts meaningful, interpretable concepts, letting you see what features the model detects. Researchers and engineers working with vision models would use this to explain and debug their models.

127 stars.

Use this if you are developing or studying large vision models and need to extract and visualize the underlying visual concepts the model has learned.

Not ideal if you are looking for a general-purpose machine learning library or a tool for natural language processing models, as it's specifically for vision-based sparse autoencoders.

AI explainability computer vision research deep learning interpretability model debugging representation learning
No Package No Dependents
Maintenance 6 / 25
Adoption 10 / 25
Maturity 16 / 25
Community 7 / 25

How are scores calculated?

Stars

127

Forks

5

Language

Python

License

MIT

Last pushed

Dec 04, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/KempnerInstitute/overcomplete"

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