SkyeLu

Has any one solved it? I met the same problem.

Thank you! Looking forward to it!

Hi, it is normal. Some of our testing results might not be quite satisfying, especially on the hunman faces (need improvements in future work).

Are you testing on your own data? If so, the LR and Ref image should be padded if they are not multiples of a number (8/16/32..., depending on the value...

For the fair comparison. Actually, there are some recent papers that propose new datasets, for example: Dual-Camera Super-Resolution with Aligned Attention Modules.

The input size only needs to be a multiple of 16.

Hi, did you stitch 4 references to one image while testing? If not, please refer to this issue: #2 You should try the following command: `python test.py --resume './pretrained_weights/masa_rec.pth' --testset...

In the eval phase during training, we do not use four reference images, only one reference image is used for fast evaluation.

> Hi, I also try to reproduce the result. The best PSNR of CUFED is only 27.294 (for L1 and 27.466 for all references) in the 250 epochs. Hi, Different...

Ref分辨率高细节多，而LR分辨率低，两者之间实际存在domain gap，直接匹配不是很合理。把Ref进行降采样也变成低分辨率之后一定程度上可以减小这种gap。另一方面，Ref降采样后匹配也能减小些计算量。