config.gin

run.batch_size = 64
run.epochs = 20
run.val_split = 0.15
run.num_workers = 32
run.print_every = 100

run.trainval_csv_path = 'merged_combined_30_sept.csv'
# run.trainval_csv_path = 'trainset_with_normal_30_sept.csv'
# run.trainval_csv_path = 'verified_oiscapture_trained_labels_with_normal.csv'
# run.test_csv_path = 'testset_filtered.csv'
# run.test_csv_path = 'testset.csv'
# run.test_csv_path = 'extra_normal_unclassified.csv'
# run.trainval_csv_path = 'trainset.csv'
# trainval_csv_path = 'trainset_with_normal.csv'
test_csv_path = 'trainset_filtered.csv'
# trainval_csv_path = 'self-training-set_filtered.csv'
# run.trainval_csv_path = 'self-training_images.csv'
run.tasks = [0, 1, 2]

run.lr = 1e-3
run.weight_decay = 1e-6
run.momentum = 0.9
run.dataset_dir = '/data2/fundus_images/'
run.model_type = 'resnet50'

MultiTaskModel.in_feats = 2048 # 1024 -> densenet121, 2048 -> resnet50, 512 -> resnet34