You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello, we were using 001_classicalSR_DIV2K_s48w8_SwinIR-M_x2.pth pretrained weight and tested it on manga-109 but the result we got is much worst than one reported in the paper.
Hello, we were using 001_classicalSR_DIV2K_s48w8_SwinIR-M_x2.pth pretrained weight and tested it on manga-109 but the result we got is much worst than one reported in the paper.
Our result: psnr: 29.23; ssmi: 0.9387
Reported: psnr: 39.6, ssmi: 0.9792
We test on on both torch 1.10 and 1.8 and the results we received are the same.
So how to reproduce the inference accuracy with the pretrained models
The text was updated successfully, but these errors were encountered: