Jianyi Wang
Jianyi Wang
Hi. Of course, it can. You need to prepare label file and corresponding image data. You may refer to the format of Koniq-10k.
@Svytta Hi~ thanks for your interest in StableSR. I am not sure what happens to huggingface though I restart the demo. These free demos are usually unstable, unfortunately. I suggest...
You can first upscale to 512 and resize to 256. Directly getting 256 will lead to poor quality since the diffusion prior is trained on 512.
Hi. I did not meet such a problem. You may check if the output is None. If so, you may need to avoid using autocasting for the autoencoder part. This...
Hi. Sry that I do not quite get your question here. The images you get are already 4x upsampled because our model takes the LR images with the same resolution...
Hi. It seems like to be a network issue. Usually retrying for several times or unsettling proxy.
> @IceClear @pkuliyi2015 Hi~ thanks for your interest and contribution of our work. The originality of this part should come from @pkuliyi2015 and is also included in our latest [IJCV...