Gradient Reversal on data of a specific class

[ptrblck]

@louvere_sam

However, my domain classifier only operates on data corresponding to the positive class.
My question is whether the domain loss and binary loss given in my below code actually affect the feature extractor?

I’m unsure if I misunderstand the question, but since domain_loss.backward() will only be called if label == 1, it won’t have any effect on the feature_extractor in other cases.Let me know, if I misunderstood the question.