AI Paper Atlas

Research map / Image generation / Image restoration

Towards Robust Blind Face Restoration with Codebook Lookup Transformer

Shangchen Zhou, Kelvin C.K. Chan, Chongyi Li, Chen Change Loy

NeurIPS 2022 · Image generation › Image restoration · approach: Efficient restoration architectures

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 the learned discrete codebook (vs plain encoder-decoder) (ablation)CodeFormer · CelebA-TestLPIPS: 0.420 -> 0.397Improved
add the codebook (vs plain encoder-decoder) (ablation)CodeFormer · CelebA-TestIDS: 0.47 -> 0.51Improved
predict codes with a linear/CNN head instead of nearest-neighbour matching (ablation)CodeFormer · CelebA-TestLPIPS: 0.397 -> 0.351Improved
use Transformer for code prediction instead of CNN (ablation)CodeFormer · CelebA-TestLPIPS: 0.351 -> 0.307Improved
use Transformer for code prediction instead of CNN (ablation)CodeFormer · CelebA-TestIDS: 0.55 -> 0.58Improved
fine-tune the decoder instead of keeping it fixed (ablation)CodeFormer · CelebA-TestLPIPS: 0.307 -> 0.379Hurt
set CFT coefficient w=1 instead of w=0 (ablation)CodeFormer · CelebA-TestLPIPS: 0.307 -> 0.297Improved
set CFT coefficient w=1 instead of w=0 (ablation)CodeFormer · CelebA-TestIDS: 0.58 -> 0.60Improved
increase codebook size N from 384 to 1024 (ablation)codebook reconstruction (Stage I) · FFHQLPIPS: 0.202 -> 0.175Improved
increase codebook size N from 384 to 1024 (ablation)codebook reconstruction (Stage I) · FFHQPSNR: 22.59 -> 23.41Improved
use multi-scale encoder features as Transformer input instead of single-scale (ablation)CodeFormer · CelebA-Testcode prediction accuracy: 0.192 -> 0.188No clear effect
increase Transformer depth from 9 to 15 layers (ablation)CodeFormer · CelebA-Testcode prediction accuracy: 0.192 -> 0.193No clear effect

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