search

MIC-DKFZ / nnUNet

Apache License 2.0
5.79k stars 1.74k forks source link

Why the training loss of CT-segmentation is negative? #2018

Closed ayf-9797 closed 7 months ago

ayf-9797 commented 7 months ago

I have experienced all negative values for training loss and validation loss when training with both 3d-full and 2d modes. However, all preprocessing operations were done all according to the instructions. What is the cause of this situation and will it affect the prediction results of the network? such as: image

ayf-9797 commented 7 months ago

Thanks to the author for maintaining this section, I have found the answer I was expecting.