[StableDiffusionInpaintPipeline] accept tensors for init and mask image

[patil-suraj]

This PR updates StableDiffusionInpaintPipeline to accept both torch.FloatTensor and PIL.Image.Image for init_image and mask_image.

Fixes #370

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[Inkorak]

@patil-suraj There are problem here, in my opinion. If we send a FloatTensor, then there will be no preprocessing and the mask variable will not be declared, and secondly, there will be no transfer to the device.

if not isinstance(mask_image, torch.FloatTensor):
   mask = preprocess_mask(mask_image).to(self.device)
mask = torch.cat([mask] * batch_size)

With this, there should be no problems:

if not isinstance(mask_image, torch.FloatTensor):
   mask_image = preprocess_mask(mask_image)
mask = torch.cat([mask_image.to(self.device)] * batch_size)

[patil-suraj]

good catch! Updating it now

[pcuenca]

LGTM!

Question: if the user provides an image and a mask as tensors, should we verify that the number of channels match? The image should have R,G,B while the mask is only L. Is this something we should check or is it too much?

On second thought, we could just update the documentation and not the code:

mask_image (`torch.FloatTensor` or `PIL.Image.Image`):
    `Image`, or tensor representing an image batch, to mask `init_image`. White pixels in the mask will be
    replaced by noise and therefore repainted, while black pixels will be preserved. If `mask_image` is a
    PIL image, it will be converted to a single channel (luminance) before use. If it's a tensor, it should
    contain one color channel (L) instead of 3, so the expected shape would be `(B, H, W, 1)`.

[patil-suraj]

Good point @pcuenca ! The pipeline is experimental and will be soon updated so just updated the docs for now.