Neuropathology Vision Transformer using DINOv2-with-registers architecture. Key innovation is the hierarchically connected tiles strategy, which improves representation learning from whole slide images.
| Metric | Value |
|---|---|
| Linear Probe Accuracy | 84.51% |
| KNN Accuracy | 87.20% |
Outperforms UNI2 (77.5%), prov-gigapath (77.0%), UNI (76.7%) on the same dataset.
Available on Hugging Face.
Hosted on Hugging Face.