nan for train_loss and val_loss during training

[laure0406]

Hello everyone !

I am currently trying to use nnUNet to segment some MRIs. I have 260 of them, with their asociated labels.
When I run the training, with nnUNetv2_train 100 3d_fullres 0 -tr nnUNetTrainer_250epochs , (after running the preprocess), from the 1st epoch, i have train_loss nan and val_loss nan.
I don't know where the problem comes from. I checked the documentation my files seem in order (json file, organisation and name of folder...)
If someone could help me I would be very grateful !!
Don't hesitate to ask for more information if you need some to understand my issue.

Thank you so much,
Have a nice day !!

PS: here is what i used:

----------- Pre-process ----------
nnUNetv2_plan_and_preprocess -d 100 -c 3d_fullres --verify_dataset_integrity -np 2

----------- Training ----------
nnUNetv2_train 100 3d_fullres 0 -tr nnUNetTrainer_250epochs