Nice code. However, I still have a question about this model. In the A phase, I found that it uses same labels but different domains dataset . However, in another way, is it uses some same label datas to augment the few shot data? Is it normal in the few-shot learning?