psnr 31.23 db of pretrained model on Set5 x4 Super Resolution about hpinet

[Davorking]

PS: after using .bmp extension, the x4 Set5 psnr result drops further to 29.93db

Namespace(model='M', scale=4, is_y=True)
#Params of HPINet-M: 895.6590 [K]
loading: checkpoints/HPINet-M-x4.pth
Set5, Mean PSNR: 29.93, SSIM: 0.8434
Set14, Mean PSNR: 28.19, SSIM: 0.7939
B100, Mean PSNR: 27.35, SSIM: 0.7493
Urban100, Mean PSNR: 26.10, SSIM: 0.7950

[Davorking]

After check my code again, and did some more inspections, I found the issue is caused by myself changing interpolation methods to "bicubic" instead of "bilinear", which is the way the model was trained.

The result now looks like this:
Namespace(is_y=True, model='M', scale=4)
#Params of HPINet-M: 895.6590 [K]
loading: checkpoints/HPINet-M-x4.pth
Set5, Mean PSNR: 32.60, SSIM: 0.8987
Set14, Mean PSNR: 28.87, SSIM: 0.7874
B100, Mean PSNR: 27.74, SSIM: 0.7421
Urban100, Mean PSNR: 26.72, SSIM: 0.8044

Which matches exactly the number listed in the paper.

Sorry for the trouble, and hope you have a nice day!