search

sefibk / KernelGAN

Other
337 stars 77 forks source link

The reproduced experimental result are lower than that in the paper. #34

Open SanSLee opened 4 years ago

SanSLee commented 4 years ago

Thanks for the great job. I reproduced the experiments for scale x2 and x4 on DIV2KRK dataset that you provided, but the PSNR/SSIM are 29.93/0.8548 and 26.76/0.7302 respectively, which are all lower than results in your paper. Could you please upload the all the visual results ?

sefibk commented 4 years ago

Sorry I do not have access to the results since I am no longer in the institute. If necessary we can try to get them. Note:

  1. PSNR code calculations vary. We used the Matlab code from VDSR if I recall. That could be one cause for the difference.
  2. KernelGAN re-trains for every image hence results vary. The reported results are the average of 5 independent runs