Summing the losses should work as long as each loss contains as valid .grad_fn as an attribute.
The code looks generally alright. I haven’t checked the architecture (e.g. if the slicing of the vgg model is correct), but the workflow looks good.
Summing the losses should work as long as each loss contains as valid .grad_fn as an attribute.
The code looks generally alright. I haven’t checked the architecture (e.g. if the slicing of the vgg model is correct), but the workflow looks good.