Continuous Magnification Sampling

This repository contains the accompanying code for our work on continuous magnification sampling in pathology foundation models, along with instructions to obtain and evaluate the benchmark data (TCGA-MS, BRACS-MS).

Note: This repository is under active development and will be updated continuously.

Release Notes

2025-01-09

• Added code to create and evaluate the BRACS-MS dataset

Getting Started

Prerequisites

• Python 3.x
• Hugging Face account (for accessing certain models)
• Installed packages from requirements.txt (pip install -r requirements.txt)

Authentication

Some public models require Hugging Face authentication. Log in before running evaluations:

huggingface-cli login --token YOUR_TOKEN_HERE

BRACS-MS Dataset

1. Download Source Data

Download the BRACS ROIs from the official source: https://www.bracs.icar.cnr.it/download/

2. Create the Dataset

Adjust the source and output paths in create_bracs_dataset.sh, then run:

./create_bracs_dataset.sh

3. Run Evaluation

./run_eval.sh

TCGA-MS Dataset

Coming soon

Citation

Coming soon

License

Coming soon