idinvert_pytorch
idinvert_pytorch copied to clipboard
Semantic Diffusion : Irregular foreground with aribitray coordinates
Is there a way to modify the source code to blend skin defects (say a birthmark on to a face) that are non-rectangular and on arbitrary coordinates on the face. Could you provide any high level thoughts parts of the code that would change?
would I have to write a version of diffuse in utils/inverter.py that accepts a mask instead of x/y coordinates of the rectangular foreground bounding box?
Yes
would I have to write a version of diffuse in utils/inverter.py that accepts a mask instead of x/y coordinates of the rectangular foreground bounding box?