search

codeslake / IFAN

[CVPR 2021] Official PyTorch Implementation for "Iterative Filter Adaptive Network for Single Image Defocus Deblurring"
GNU Affero General Public License v3.0
227 stars 39 forks source link

About th evalaution on DPDD #1

Closed xianrui-luo closed 3 years ago

xianrui-luo commented 3 years ago

Hi, I notice that in your paper, the test results of DPDNet is different from the original paper, where in the original paper PSNR is 25.13,and in your paper PSNR is 25.23. I wonder what causes the difference?

codeslake commented 3 years ago

Hi!

PSNR in the main paper of DPDNet is measured with the DPDNet trained and tested with 16-bit images, but all the models we've compared in our paper are trained and tested with 8-bit images.

The PSNRs that we have reported in our paper are measured from the checkpoint of DPDNet(_dual) trained with 8-bit images that we acquired from the authors. But it is still different from what they have reported in their supplementary material (Table 5). We just reported higher PSNR in our paper.

Also, since the authors of DPDNet did not have the checkpoint of DPDNet(_single) trained with 8-bit images, we've trained the network with the training code provided by the authors.

Thanks

xianrui-luo commented 3 years ago

Thank you for replying!