Pad images to size requirements in call API

chaiNNer-org / spandrel

Spandrel gives your project support for various PyTorch architectures meant for AI Super-Resolution, restoration, and inpainting. Based on the model support implemented in chaiNNer.

MIT License

105 stars 7 forks source link

Pad images to size requirements in call API #137

Closed RunDevelopment closed 5 months ago

RunDevelopment commented 5 months ago

Changes:

Add get_padding for SizeRequirements. This method will return the necessary to make a size satisfy the requirements.
Automatically pad (and later unpad) input images in the call API. This means that user code no longer has to handle padding and can run models on images of any size.
Some additional checks to verify call API requirements.

RunDevelopment commented 5 months ago

Getting padding right turned out to be more interesting than I thought. The issue is that torch.nn.functional.pad(t, (0, pad_w, 0, pad_h), "reflect") only works if pad_w and pad_h are less than the width and height of the image respectively. So we first have to pad as much as we can with reflect and then use something else to cover the rest (if anything). Surprisingly non-trivial.