AI Paper Atlas

Research map / Image generation / Image restoration

DifFace: Blind Face Restoration With Diffused Error Contraction

Zongsheng Yue, Chen Change Loy

TPAMI 2024 · Image generation › Image restoration · approach: Priors for restoration networks · 136 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add diffusion prior to SwinIR estimator (ablation)SwinIR · CelebA-Test BFRFID-G: 93.73 -> 20.29Improved
add diffusion prior to SwinIR estimator (ablation)SwinIR · CelebA-Test BFRPSNR: 24.32 -> 23.44Hurt
add diffusion prior to SRCNN estimator (ablation)SRCNN · CelebA-Test BFRFID-G: 108.68 -> 44.83Improved
use SwinIR instead of SRCNN as diffused estimator (ablation)DifFace · CelebA-Test BFRFID-G: 44.83 -> 20.29Improved
increase DDIM eta from 0 to 1.0 (ablation)DifFace · CelebA-Test BFRLPIPS: 0.475 -> 0.443Improved
increase DDIM eta from 0 to 1.0 (ablation)DifFace · CelebA-Test BFRPSNR: 23.49 -> 23.28Hurt
increase DDIM eta from 0 to 0.75 (ablation)DifFace · CelebA-Test BFRFID-G: 23.51 -> 19.04Improved
reduce sampling steps from 250 to 20 (N=100 to 8) (ablation)DifFace · CelebA-Test BFRFID-G: 20.29 -> 42.54Hurt
reduce sampling steps from 250 to 50 (N=100 to 20) (ablation)DifFace · CelebA-Test BFRFID-F: 48.98 -> 54.32Hurt
reduce sampling steps from 250 to 100 (N=100 to 40) (ablation)DifFace · CelebA-Test BFRPSNR: 23.44 -> 23.41No clear effect
optimization step with gamma > 0 for inpainting (ablation)DifFace · CelebA-HQ inpaintingcoherenceImproved
use diffusion-based method vs GAN-based for inpaintingDifFace · CelebA-HQ inpaintingFID-G: 34.73 -> 24.45Improved

Later papers that compared with this method

9 later accepted papers report DifFace as a baseline in their results tables; 8 of them report a better number on most of the results they share with it.

1 more not listed.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy