Custom module loss containing multiple losses

[EomSooHwan]

Hello, I want to try Text Recognition with SVTR and CRNN with customized module loss. In detail, I am trying to apply custom regularization on CTC. So, the total loss for the model will be total_loss = loss_ctc + regularization_weight * loss_regularization.

I haven't run the code yet, but I'd like to know if there is a way to train with this custom loss while being able to visualize the loss curve for each loss_ctc and loss_regularization. Will losses = dict(loss_ctc = loss_ctc, loss_regularization = regularization_weight * loss_regularization) do a trick? Or, for example, will changing into losses = dict(total_loss = total_loss, loss_ctc = loss_ctc.detach(), loss_regularization = loss_regularization.detach()) still work properly?

[EomSooHwan]

I think this does the trick for now.