Details about how to process data before training

[YiBaiHP]

Hi! I am trying to reproduce your good work to go on multi-modal registration between CT and MR images. But i find the output always has NAN, could you tell me how can I solve the problem? Should I do some data process before training?
Thanks for your attention!

[YiBaiHP]

My label only have 4 categories, which does not consider the background. Could you also tell me how to change the code to get a better performance? Thank you very much.