search

wyhuai / DDNM

[ICLR 2023 Oral] Zero-Shot Image Restoration Using Denoising Diffusion Null-Space Model
MIT License
1.08k stars 75 forks source link

nderstanding Discrepancies: Challenges in Achieving Comparable Inpainting Results to Published Papers on ImageNet Dataset #39

Open hosseinaskari-cs opened 1 year ago

hosseinaskari-cs commented 1 year ago

Why am I unable to achieve the same high-quality results as the paper when performing the inpainting task on the ImageNet dataset?

hosseinaskari-cs commented 1 year ago

For example, the paper reported a PSNR of 32.06 in the inpainting task for imagenet dataset, but I am getting only 23.29.

wyhuai commented 1 year ago

It depends on what kind of mask you use.

In our paper, we use the same mask generation code as DDRM for a fair comparison with them. But in this repository, we only provide one mask for visual evaluation, which is not the mask used in quantitative comparisons. You can change the mask as you will.

image

hosseinaskari-cs commented 1 year ago

Thank you so much for your swift reply. I am trying to modify your method for other applications. I would like to know if we can have your point of view. Your contribution will be highly appreciated.