The Pathology AI Library

The library

UNI

Self-supervised vision foundation model for pathology tile encoding.

Cost
Free for academic use
Hardware
Needs a consumer GPU
Offline
Works without a connection
Scanner
Not required
Your data
Stays on your machine
Licence
gated
Maturity
production
From
Mahmood Lab, Harvard Medical School
Model weightsSource codePaper

What it is

ViT encoder trained with DINOv2 on a large private histopathology corpus, used as a frozen tile-level feature extractor across a wide range of downstream tasks.

Caveats

Gated weights with a non-commercial licence. Trained on a corpus that is not public, so independent replication is limited.

Licence notes

Access is gated on Hugging Face and the licence restricts commercial use. Read the terms before any non-research application.

Activity: 763 stars · last commit 2025-03-26 · 101,153 downloads — refreshed nightly

Related

clamtridenttitan

Catalogued 2026-08-01, last checked 2026-08-01. View the source record · Report an error