Closed GarlandZhang closed 4 years ago
One observation is we are only interested in the hair style for the sides and back... any other facial features only matter for the frontal view. So we can only apply attributes that are hair-related.
Now all we need to know is when to apply this restriction.
What happens if we feed a headshot image into the gan but its not frontal facing? In particular, suppose we had a photo of the head slightly sideways (but the face is mostly still in view). Does it accidentally add a mustache to the side of the head, too? Or does it somehow know?
This idea might be too ambitious
We want to run this model so video output (whether it is a video recording, or real time) gives the translated form. Since we only trained on faces from the frontal perspective, we do not have the flexibility to check the sides or back. This limits the robustness of our model.