AI Paper Atlas

Research map / Image generation / Image restoration

RestoreFormer: High-Quality Blind Face Restoration From Undegraded Key-Value Pairs

Zhouxia Wang, Jiawei Zhang, Runjian Chen, Wenping Wang, Ping Luo

CVPR 2022 · Image generation › Image restoration · approach: Priors for restoration networks · 125 citations (OpenAlex)

Check 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
use self-attention on degraded-image features instead of no attention (ablation)RestoreFormer · CelebA-TestFID: 50.68 -> 47.39Improved
use self-attention on degraded-image features instead of no attention (ablation)RestoreFormer · CelebA-TestIDD: 0.6401 -> 0.6284Improved
replace MHCA with local SFT fusion (ablation)RestoreFormer · CelebA-TestFID: 41.45 -> 41.47Hurt
replace MHCA with local SFT fusion (ablation)RestoreFormer · CelebA-TestIDD: 0.5650 -> 0.6702Hurt
use only degraded-image features with MHSA (ablation)RestoreFormer · CelebA-TestIDD: 0.5650 -> 0.6284Improved
use only dictionary priors with MHSA (ablation)RestoreFormer · CelebA-TestFID: 41.45 -> 45.83Improved
use only dictionary priors with MHSA (ablation)RestoreFormer · CelebA-TestIDD: 0.5650 -> 0.7662Hurt
fuse degraded-image and prior information with MHCA rather than using either source alone (ablation)RestoreFormer · CelebA-TestFID: 45.83 -> 41.45Improved
fuse degraded-image and prior information with MHCA rather than using either source alone (ablation)RestoreFormer · CelebA-TestIDD: 0.6284 -> 0.5650Improved
use degraded features rather than priors in the MHCA skip connection (ablation)RestoreFormer · CelebA-TestFID: 41.45 -> 42.00Hurt
use degraded features rather than priors in the MHCA skip connection (ablation)RestoreFormer · CelebA-TestIDD: 0.5650 -> 0.5938Improved
replace the reconstruction-oriented dictionary with a recognition-oriented VGG dictionary (ablation)RestoreFormer · CelebA-TestFID: 41.45 -> 61.43Hurt

Later papers that compared with this method

7 later accepted papers report RestoreFormer++ as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 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