Dice coefficient is lower than the value published in the paper.

[chaojuncheung]

I trained directly with "python train.py", and reduced the batch size to 8. My final result is only 71.9 with this setting, did you modify the parameters during the segmentation training? Thank you.

[chaojuncheung]

Hello author, I have another question. I directly loaded the checkpoint you provided, and the mean dice obtained in the test is also around 72. Does the code make a difference when evaluating the segmentation results? If it is different, can you provide the code at the time of evaluating? Many thanks! !