What loss functions would you choose for an image-to-image translation task, and how do you balance their weights?
