Open zzzpc opened 3 years ago
In my view, the target encoder aims to map the target features into the source domain so that we can use the pre-trained source classifier to classify them. Therefore, we would like to make the target label after discriminating close to the source label that is one.
cause in line 63 label_tgt = make_variable(torch.zeros(feat_tgt.size(0)).long()) why in line 91 label_tgt changes to torch.ones(feat_tgt.size(0)).long() ?