Research map / Image generation / Image restoration
TPCNet: A Low-Light Image Enhancement Network Inspired by Triple Physical Constraints
Jing-Yi Shi, Ming-Fei Li, Ling-An Wu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add CAM (color-association mechanism) to Baseline-1 (ablation) | DCGT baseline · LOL-v2-Real | PSNR: 21.637 -> 23.729 | Improved |
| add LFE and RFE estimators on top of CAM (ablation) | Baseline-1 + CAM · LOL-v2-Real | PSNR: 23.729 -> 24.357 | Improved |
| add the full triple physical constraints (Eqs. 3-5) (ablation) | TPCNet w/o TPC · LOL-v2-Real | PSNR: 24.357 -> 24.641 | Improved |
| add Eq. (3) constraint only (ablation) | TPCNet w/o TPC · LOL-v2-Real | PSNR: 24.357 -> 24.598 | Improved |
| add Eqs. (4-5) only (ablation) | TPCNet w/o TPC · LOL-v2-Real | PSNR: 24.357 -> 24.575 | Improved |
| enforce Eq. (5) by explicit division (ablation) | TPCNet · LOL-v2-Real | PSNR: 24.357 -> 24.188 | Hurt |
| learn alpha instead of fixing alpha = 1 (keep specular term) (ablation) | TPCNet · LOL-v2-Real | PSNR: 24.523 -> 24.641 | Improved |
| use YCbCr instead of HVI or LAB in CAM (ablation) | TPCNet · LOL-v2-Real | PSNR: 24.641 (HVI) / 24.588 (LAB) -> 24.978 | Improved |
| construct physical constraints in feature space (TPCNet) vs image space | TPCNet vs Retinexformer · LOL-v2-Real -> LOL-v2-Synthetic | PSNR: 13.998 -> 17.058 | Improved |
| TPCNet vs HVI-CIDNet | TPCNet · LOL-v2-Real | PSNR: 24.111 -> 24.978 | Improved |
| TPCNet vs QuadPrior on unpaired data | TPCNet · MEF/NPE/LIME/DICM/VV | MUSIQ: 62.241 (QuadPrior) vs 58.070 | Mixed |
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.
- Retinexformer — Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement (ICCV 2023) · better on 5 of 5 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
- Retinex-Inspired Unrolling With Cooperative Prior Architecture Search for Low-Light Image Enhancement (CVPR 2021 · similarity 0.87)
- Learning a Simple Low-Light Image Enhancer From Paired Low-Light Instances (CVPR 2023 · similarity 0.87)
- IniRetinex: Rethinking Retinex-type Low-Light Image Enhancer via Initialization Perspective (AAAI 2025 · similarity 0.86)
- Specularity Factorization for Low-Light Enhancement (CVPR 2024 · similarity 0.85)
- MambaLLIE: Implicit Retinex-Aware Low Light Enhancement with Global-then-Local State Space (NeurIPS 2024 · similarity 0.85)
- Zero-Shot Low-Light Image Enhancement via Latent Diffusion Models (AAAI 2025 · similarity 0.85)
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.
