afiosman / dense-dilated-residual-convolutions-for-contrast-enhanced-mri-synthesis Goto Github PK

Thanks for your excellent work! I am trying to reproduce the result. And I found that there are a pretrained model loading process. The upload repository didn't contain it. So I annotated these code and run the model. And I'm using the same BRATS2021 dataset. But the results is very poor, and different from the papaer result. The prediction result shows as below, The prediction is pretty white and differs greatly from the true value.

The training process and loss curve are list below:

Why this happen? How could I reproduce the paper result performance? It will be very grateful that if you could help me solve this problem.