Hi author! Thanks for your work, it's very impressive and inspiring. My question is, how does the generated image keep some attributes of the target image, like pose and expression.
From the results, some low level attributes are replaced by the source image(which provides the style), like the hair, lighting and colors, face id and age(fused with the target a little), but the head pose and face expression are kept consistent with the target. so I wonder which part of the network works for the consistency, why are they not replaced by the style, thanks a lot!
Hi author! Thanks for your work, it's very impressive and inspiring. My question is, how does the generated image keep some attributes of the target image, like pose and expression. From the results, some low level attributes are replaced by the source image(which provides the style), like the hair, lighting and colors, face id and age(fused with the target a little), but the head pose and face expression are kept consistent with the target. so I wonder which part of the network works for the consistency, why are they not replaced by the style, thanks a lot!