wrong datasets

[linshuijin6]

thanks for your sharing of your work, I find some inspiration.
While I found the psnr of results unstable.
when I look into the datasets you offered, it seems that many of them meaningless like below ( I transform the .mat file into .png for insights) , did you make more washing before infer? or if I'm wrong, please tell me, thanks a lot!

[XiaJiayu22]

inference算psnr：
y = np.nonzero(SPETimg) # 取非黑色部分
SPETimg_1 = SPETimg[y]
EPETimg_1 = EPETimg[y]
IPETimg_1 = IPETimg[y]
然后再用这部分去算psnr，ssim