Hello, thank you for releasing the excellent work and code.
I encountered a small issue while reproducing the results using your provided pretrained model.
In the paper, the reported performance on the GoPro test set is:
PSNR: 33.91
SSIM: 0.968
Using the released pretrained weights, I obtained:
Average PSNR: 33.9072 dB
Average SSIM: 0.9491
The PSNR is almost identical to the reported result, but the SSIM is noticeably lower. Therefore, I suspect the model output itself is correct, and the difference may come from the evaluation protocol or SSIM calculation method.
Could you please clarify:
Whether SSIM is computed on the RGB image or the Y channel?
Whether any crop border or additional post-processing was applied during evaluation?
I would really appreciate any suggestions or clarification.
Thank you very much for your time and help!
Hello, thank you for releasing the excellent work and code.
I encountered a small issue while reproducing the results using your provided pretrained model.
In the paper, the reported performance on the GoPro test set is:
PSNR: 33.91
SSIM: 0.968
Using the released pretrained weights, I obtained:
Average PSNR: 33.9072 dB
Average SSIM: 0.9491
The PSNR is almost identical to the reported result, but the SSIM is noticeably lower. Therefore, I suspect the model output itself is correct, and the difference may come from the evaluation protocol or SSIM calculation method.
Could you please clarify:
Whether SSIM is computed on the RGB image or the Y channel?
Whether any crop border or additional post-processing was applied during evaluation?
I would really appreciate any suggestions or clarification.
Thank you very much for your time and help!