search

sun-umn / DMPlug

This is the official implementation of "DMPlug: A Plug-in Method for Solving Inverse Problems with Diffusion Models" (NeurIPS 2024).
36 stars 3 forks source link

Question about table results and also dataset #3

Open hosseinaskari-cs opened 2 days ago

hosseinaskari-cs commented 2 days ago

Thank you for your excellent work—congratulations!
I have two questions I hope you can help me with:

Baseline Results Confusion: In Table 1, the super-resolution result for the CelebA dataset using the resampling algorithm shows an LPIPS score of 0.080. However, in the original paper, the reported result is 0.144. Could you clarify this discrepancy?

Accessing the CelebA Test Dataset: How can I obtain the CelebA test dataset in PNG format—the same one you and Resample used?

Best, Hossein

HengkangWang commented 2 days ago

Hi @hosseinaskari-cs, thank you very much for your interest in our work!

RE Baseline Results Confusion: All LPIPS scores reported in this paper are based on the default backbone, i.e., AlexNet. Scores may differ if alternative backbones are used.

RE Accessing the CelebA Test Dataset: We first download the CelebA dataset from the official website https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html and then resize the images to 256 * 256 to match the resolution of DDPM/DDIM models.

Please feel free to ask if you have further questions.

hosseinaskari-cs commented 8 hours ago

Thank you so much for your reply.

Is there any chance I have only your test dataset for Celebas? This would help me a lot.