SSIM is weirdly high in the paper

Question

SSIM is weirdly high in the paper

guochengqian opened this issue 2 years ago · comments

Gordon Guocheng Qian 钱国成 commented 2 years ago

Dear authors,
Thank you for your work.

I have run your pretrained weights (JDnDmSR+) and tested PSNR and SSIM using the standard MATLAB code.
Here is what I get:
PSNR: 26.88, SSIM: 0.8307

the PSNR achieves a close value to the paper (26.89), but the SSIM is way lower than paper (0.8922). I guess you measured SSIM using some other implementation that might not be the same as used in the literature.

Could you comment something on this?

xingwz · Answer 1 · Mon Jun 13 2022 00:49:14 GMT+0800 (China Standard Time)

We just use the sim function of Matlab,
ssim = ssim(imageTest, imageGT, 'DynamicRange', 255);
Please we cut the border (=sf) of the images before the evaluation.

Gordon Guocheng Qian 钱国成 · Answer 2 · Mon Jun 13 2022 01:34:51 GMT+0800 (China Standard Time)

Hi @xingwz thank you for the clarification.