In the process of training, I encountered a problem of excessive loss

[Yuri-delta]

Hi, I'm using the u-net, my_cross_entropy2d in training. But the loss is too excessive as follows. Then I use the model to test. The gray value of the output label is all 0. Could you tell me how to deal with this problem？

[tenres]

Hi Yuri-delta, were you able to resolve the above problem? Thanks!