search

MC-E / Deep-Generalized-Unfolding-Networks-for-Image-Restoration

Accepted by CVPR 2022
123 stars 26 forks source link

I don't find the result of DGUNet on the DND benchmark #1

Closed IDKiro closed 2 years ago

IDKiro commented 2 years ago

It's an amazing work! Training the denoising network to 40.12dB using only the SIDD dataset seems to be the best current performance. I don't understand why the results are hidden on DND.

IDKiro commented 2 years ago

I am very interested in DGUNet and want to follow it because of its strong performance and godd interpretability. However, FGDM does not seem to be suitable for image denoising because the noise degradation process is not deterministic (not Ax but + n)?

MC-E commented 2 years ago

Thanks for your attention and valuable suggestion! We have updated the evaluation for DND. In real-world scenarios, image denoising is not only related to additive noise (i.e., A!=I).

IDKiro commented 2 years ago

Thanks for your reply!