Unable to reproduce CHASEDB1 results

[apuomline]

Hello author, I renamed the CHASEDB1 dataset and used the 1st mask image for training. I followed your training configuration and the final result is as follows:

However, in the paper, the results of the data set are as follows:

The following is my configuration of the training settings and the division of the training set and the validation set:

[Jiaoyang45]

The difference between the configuration used in the paper and your configuration is as follows:

1. Img_ext=. jpg.
2. Lr=0.001, weight_decay=0.001.  Also, please note:
3. The "test_size" of "train_test_split" is not 0.2, but 0.27, which makes the training set exactly 20 images, which is a multiple of batch_size, in order to fully utilize the data for learning.
4. The "random_state" of "train_test_split" are 41, 1029, and 3407, respectively. The 'seed' of the 'seed_torch' is 1029.