about softmax_loss layer

[qinhaifangpku]

hi~ It is very appreciated that you release this code. It is very helpful for my research. but I have the question about the softmax_loss layer. in some situation that the batch data has just C class which C < K, K is the total class in the networks. and it will LOG error that: "sum of pixel-wise loss weights is zero" Do you have ever happened to this? Is it matter with the results?

thank you very much in advance!

[holgerroth]

Yes, it happens if all voxels within the patch have a zero weight or are labelled with the ignore value. If you use the weight image, they should be non-zero somewhere.

[qinhaifangpku]

but if we have two class to classified, the patches for training sometimes just have class 1, so the weights of class 2 will be zero. because the samples are imbalance, it is not possible to make all the patches for training contain two class label at the same time.

[sara-eb]

@qinhaifangpku Could you solve the sum of pixel-wise loss weights is zero issue? If yes, could you please share it here? The loss that i am getting in some points becomes negative, which I think this warning should be the reason. Your help is appreciated.