Diabetic Retinopathy Detection with AI

Setup

Cloning the repo

git clone https://github.com/SDAIA-KAUST-AI/diabetic-retinopathy-detection.git
git lfs install # to make sure LFS is enabled
git lfs pull # to bring in demo images and pretrained models

Gradio app environment

Install from pip requirements file:

conda create -y -n retinopathy_app python=3.10
conda activate retinopathy_app
pip install -r requirements.txt
python app.py

Install manually:

pip install pytorch --index-url  https://download.pytorch.org/whl/cpu
pip install gradio
pip install transformers

Training environment

Create conda environment from YAML:

mamba env create -n retinopathy_train -f environment.yml

Download the data from Kaggle or use kaggle API:

pip install kaggle
kaggle competitions download -c diabetic-retinopathy-detection
mkdir retinopathy_data/
unzip diabetic-retinopathy-detection.zip -d retinopathy_data/

Launch training:

conda activate retinopathy_train
python train.py

The trained model will be put into lightning_logs/.