Open suke27 opened 6 years ago
There is almost no difference about the structure of the image transformation networks.
you make a hidden layers when style loss backward? input is style and transfer image gram loss, output is 1792, could you explain this process, is it classify gram loss to point which style it belong?
The meta network builds a mapping from a style image to a 1792-d vector.
Hi, could you help to explain difference with Johnson's, just have small network (resblock fitler 32)?