Closed yumianhuli1 closed 10 months ago
You can try more recent models from different authors - they may be better. In our repository, big-lama is the best. All the rest models are both faster and deliver a bit worse image quality. If you know the size of your images, you can try distilling big-lama into a lighter architecture without Fourier layers (but it will not be as good at repetitive patterns without that). Another option is to use torch.compile or port it to a different framework. Neither of this is done in this repo, so it is up to you and the community
There are models that are faster than big-lama inference without much loss of accuracy ?And also have a download link thanks