pip3 install --user -r requirements.txt
- Diabethic Rethinopathy Detection (2015), both train and test subsets
- APTOS 2019 Blindness Detection (2019), dublicate images from train subset are removed, final label file: input/trainLabels19_unique.csv, dublicate removal script: preprocessing/2019_train_find_dublicates.py.
- Messidor-1, [0-4] adjusted labels, final label file: input/messidor1_labels_adjudicated.csv.
For 2015 and 2019 data, this dataset with resized images was used. Images from Messidor dataset were also resized and converted to jpeg format using preprocessing/messidor_tif_2_jpg.py.
python3 train_efficientnet.py -sid N
where N is Cross-Validation fold id, i.e. [0..4]
python3 eval_ensemble.py