I am working on style transfer and almost all papers that I checked, the authors assign weights to the content and style losses explicitly in their code. But the network also provides implicit weights to the losses by itself, if I am not wrong.
Why should we again specify the weights then?