Self-Supervised Vision Transformers for Breast Histopathology Image Embeddings in Invasive Ductal Carcinoma Detection

This project was developed by Faiaz Rahman originally for CS 482: Applied Machine Learning under Dr. David van Dijk at Yale University.

Setup

We recommend using a virtual environment via Conda. We have provided an environment YAML file to rebuild the same virtual environment used in our experiments. We use Python 3.7, PyTorch 1.11.0, and CUDA 11.3.1.

conda env create --file environment.yml
conda activate ssbh-transformers

Data

cd data
kaggle datasets download paultimothymooney/breast-histopathology-images
unzip breast-histopathology-images.zip

If you have issues with the Kaggle API, create a separate virtual environment (to be used only for data downloading) and try running as follows.

cd data
conda env create --name download-data python=3.7
conda activate download-data
pip install kaggle
python -m pip install requirements.txt
kaggle datasets download paultimothymooney/breast-histopathology-images
unzip breast-histopathology-images.zip
conda deactivate

Then, reactivate your main ssbh-transformers virtual environment to continue with running the experiments.