Using transfer learning to classify diabetic retinopathy fundus images

1. Dataset https://ieee-dataport.org/open-access/indian-diabetic-retinopathy-image-dataset-idrid

2. Dataset dir:

IDRID_Data/
    -train_image/
    -test_image/
    -train.csv
    -test.csv

3. Prepare:
   Change data_EDA directory. Run data_EDA to convert label, create TFRecord files, and show label distribution

python input_pipeline/data_EDA.py

4. Training and fine tuning:
   Change dataset directory in config.gin, set train flag to Ture in main.py

python main.py

5. Hyperparameter tuning:
   Change config.gin path to your own path

python tune.py

6. Deep Visualization:
   Change model_dir and dataset_dir to your own directory

python grad_cam.py