Thank you for this great model which performs very quickly. I notice a significant issue that could be related to rgb_mean and img_range values. When using your pretrained model spanx4_ch48.pth, upscaled image almost always contain pixel artifacts that make final picture unusable (see image) :
When I use their codes, I normalize the image as in their codes but don't do de-normalization, but clip the output to [0, 1]. By doing so, the output looks good to me.
Thank you for this great model which performs very quickly. I notice a significant issue that could be related to rgb_mean and img_range values. When using your pretrained model spanx4_ch48.pth, upscaled image almost always contain pixel artifacts that make final picture unusable (see image) :
How can I solve this problem ? Thanks.