Reproduction Issue

[YUjh0729]

Hi! Thanks for sharing your work.
I replicated the paper's code, but the results astonished me. My training and testing Dice results exceeded those in the original paper by a significant margin, and I don't know why! I used the 2D Abdomen MR dataset and training environment provided by the paper. I suspect that the training process automatically loads the authors' pre-trained model, allowing the results to achieve great performance after about 30 epochs.
Could you provide the progress.png image of your training results?