Image Feature Extraction
timm
Safetensors
pathology
histopathology
her2
breast-cancer
h-and-e
vision-transformer
multiple-instance-learning
Instructions to use cloudkites/her2-reader with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- timm
How to use cloudkites/her2-reader with timm:
import timm model = timm.create_model("hf-hub:cloudkites/her2-reader", pretrained=True) - Notebooks
- Google Colab
- Kaggle
Download data/paper/paper-data/table_pathbench.tex from cloudkites/her2-reader: direct link, hf CLI and curl.
- Browser
- Download file 966 Bytes
-
https://huggingface.co/cloudkites/her2-reader/resolve/main/data/paper/paper-data/table_pathbench.tex
- Command line
-
hf download hf://cloudkites/her2-reader/data/paper/paper-data/table_pathbench.tex
-
curl -L -o table_pathbench.tex https://huggingface.co/cloudkites/her2-reader/resolve/main/data/paper/paper-data/table_pathbench.tex
966 Bytes
| % Generated by docs/paper/paper_pathbench.py (local PathBench, HER2 status, 10 paired seeds) | |
| \begin{tabularx}{\linewidth}{@{}>{\raggedright\arraybackslash}X l r cc cc@{}} | |
| \toprule | |
| & & & \multicolumn{2}{c}{TCGA-BRCA} & \multicolumn{2}{c}{Yale-HER2} \\ | |
| \cmidrule(lr){4-5}\cmidrule(lr){6-7} | |
| tile encoder & architecture & params & AUC & $p$ vs best & AUC & $p$ vs best \\ | |
| \midrule | |
| \textbf{Ours (final)} & ViT-S/16 & 22M & \textbf{0.689$^{1}$} & -- & \textbf{0.897$^{3}$} & 0.16 \\ | |
| H-optimus-0 & ViT-g/14 & 1.1B & 0.660$^{2}$ & 0.08 & 0.913$^{1}$ & -- \\ | |
| Phikon & ViT-B/16 & 86M & 0.656$^{3}$ & 0.18 & 0.901$^{2}$ & 0.18 \\ | |
| Midnight-12k & ViT-g/14 & 1.1B & 0.630$^{4}$ & $<$0.01 & 0.878$^{4}$ & 0.02 \\ | |
| Hibou-B & ViT-B/14 & 86M & 0.605$^{5}$ & $<$0.01 & 0.853$^{5}$ & $<$0.01 \\ | |
| DINOv2-S$^\ast$ & ViT-S/14 & 22M & 0.541$^{7}$ & $<$0.01 & 0.794$^{6}$ & $<$0.01 \\ | |
| ResNet-50$^\ast$ & CNN & 26M & 0.567$^{6}$ & $<$0.01 & 0.730$^{7}$ & $<$0.01 \\ | |
| \bottomrule | |
| \end{tabularx} | |