When performing inpainting, if the provided image does not have a suitable aspect ratio that aligns with the multiple stages of the IF, the support noise and inpainting mask end up having incompatible shapes with the generated image. As a result, the assertion in base.py is triggered: assert support_noise.shape == (1, 3, image_h, image_w).
This PR offers a solution by resizing the support noise and inpainting mask appropriately to ensure their compatibility with the generated image size.
When performing inpainting, if the provided image does not have a suitable aspect ratio that aligns with the multiple stages of the IF, the support noise and inpainting mask end up having incompatible shapes with the generated image. As a result, the assertion in
base.py
is triggered:assert support_noise.shape == (1, 3, image_h, image_w)
.This PR offers a solution by resizing the support noise and inpainting mask appropriately to ensure their compatibility with the generated image size.