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.
I wanted to make review comments, but there was a lot of small things, so I just quickly did them myself.
In general, I made the code less defensive. You almost always checked whether a key was present in the state dict before accessing it. This is entirely unnecessary, because we assume that the state dicts load gets are valid. So they aren't missing any important keys. If they were missing any such keys, then they'd be invalid and we'd be allowed to just fail.
Using this assumption, I simplified some code.
Other changes:
Fixed a bug where the tag for n_blocks as create before the real value was assigned.
Removed size requirement, because it allowed all image sizes.
I wanted to make review comments, but there was a lot of small things, so I just quickly did them myself.
In general, I made the code less defensive. You almost always checked whether a key was present in the state dict before accessing it. This is entirely unnecessary, because we assume that the state dicts
loadgets are valid. So they aren't missing any important keys. If they were missing any such keys, then they'd be invalid and we'd be allowed to just fail. Using this assumption, I simplified some code.Other changes:
n_blocksas create before the real value was assigned.