dice loss, backward problem

[jasw1001]

hi there, we ran the program, and successfully train a model. but when we deeply take a look into the program, we found the program ran loss.backward which used the pytorch autograd instead of the one written in dice_loss class.

so we are wondering the reason why a backward function was written, and how to call the manually written backward function???

[hfarhidzadeh]

Hi @jasw1001 . Did you find any solution for that? It seems it goes to the customized backward function but I cant debug it. My main problem is that I cant use multi-GPU for this backward function (it jumps out without any specific error). Do you have any idea how to used torch.autograde and skip this backward function?

[jasw1001]

It seems that the torch.autograde cannot be used in Dice loss function, because some functions in Dice loss fuction are not supported by torch.autograde. For the debug problem, it's a bug in pytorch, so you may use 'ipdb' or 'pdb' to debug it.

[hfarhidzadeh]

Thanks @jasw1001 . Have you tried to implement your own version of loss function? I have used 'pdb' but it stuck in a line after #import pdb #pdb.set_trace() and never went through next lines. It is very strange for me.

[abhiML]

Hi @jasw1001 @CSMEDEEP could you give the steps as to how to start the training? The code uses some preprocessed files from the original Luna16 dataset. How do I run the preprocessing?

[491506870]

@abhiML hello~ i really want to know how to preprocess the dataset LUNA16 to get the files "normalized_brightened_CT_2_5"etc.. , have you figure it out? thanks a lot~`