AI Paper Atlas

Research map / Image generation / Image restoration

Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement

Yuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang, Radu Timofte, Yulun Zhang

ICCV 2023 · Image generation › Image restoration · approach: Low-light image enhancement · 1,086 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 one-stage Retinex framework (ORF) to the baseline (ablation)Retinexformer baseline · SDSD-outdoorPSNR: 26.47 -> 27.92Improved
add IG-MSA to the baseline (ablation)Retinexformer baseline · SDSD-outdoorPSNR: 26.47 -> 28.86Improved
combine ORF and IG-MSA (ablation)Retinexformer baseline · SDSD-outdoorPSNR: 26.47 -> 29.84Improved
estimate light-up map L_bar and multiply instead of dividing by illumination map L (ablation)ORF · SDSD-outdoorPSNR: 28.97 -> 29.26Improved
divide by estimated illumination map L (I ./ L) rather than no lighting (ablation)ORF · SDSD-outdoorPSNR: 28.86 -> 28.97No clear effect
feed light-up feature F_lu to the corruption restorer (ablation)ORF · SDSD-outdoorPSNR: 29.26 -> 29.84Improved
use IG-MSA instead of global MSA (ablation)Retinexformer · SDSD-outdoorPSNR: 28.43 -> 29.84Improved
use IG-MSA instead of window MSA (ablation)Retinexformer · SDSD-outdoorPSNR: 28.65 -> 29.84Improved
use Retinexformer instead of SNR-Netlow-light enhancement · LOL-v1PSNR: 24.61 -> 25.16Improved
use Retinexformer as preprocessing for low-light detection instead of SNR-NetYOLO-v3 on ExDark · ExDarkmean AP: 65.3 -> 66.1Improved
use Retinexformer instead of SNR-Net on LOL-v2-real SSIMlow-light enhancement · LOL-v2-realSSIM: 0.849 -> 0.840Hurt

Later papers that compared with this method

36 later accepted papers report Retinexformer as a baseline in their results tables; 28 of them report a better number on most of the results they share with it.

28 more not listed.

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