Research map / Image generation / Image restoration
Low-light image enhancement
Brighten dark images with unsupervised, contrastive, and diffusion-based methods.
72 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Image restoration, keyword cluster “removal · enhancement · color”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 3 of 6 tracked benchmarks.
24 papers in 2023–24 and 23 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 7.7% to 5.3%.
Open questions the data points to
- LOL (PSNR): best went from 27.12 (up to 2024) to 27.51 in 2025 — Low-Light Image Enhancement via Generative Perceptual Priors (AAAI 2025 · 23 citations)
- LOLv1 (PSNR): best went from 27.84 (up to 2024) to 28.2 in 2025 — HVI: A New Color Space for Low-light Image Enhancement (CVPR 2025 · 212 citations)
- LOLv1 (SSIM): best went from 0.877 (up to 2024) to 0.889 in 2025 — HVI: A New Color Space for Low-light Image Enhancement (CVPR 2025 · 212 citations)
- Papers report Pydiff on LOLv2-Syn (PSNR) anywhere from 17.33 to 29.27: comparisons against it depend on whose number you take.
What looks settled
- LOL (SSIM): best up to 2024 0.928, best in 2025–26 0.797 from 2 papers, at most 0.2% better, if at all.
- DICM (NIQE): best up to 2024 3.23, best in 2025–26 3.441 from 4 papers, at most 0.2% better, if at all.
- LOL (LPIPS): best up to 2024 0.078, best in 2025–26 0.081 from 3 papers, at most 0.2% better, if at all.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · Fast Haze Removal for Nighttime Image Using Maximum Reflectance Prior (CVPR 2017 · 206 citations)
- 2018 · Learning to See in the Dark (CVPR 2018 · 1,403 citations)
- 2019 · Underexposed Photo Enhancement Using Deep Illumination Estimation (CVPR 2019 · 997 citations)
- 2020 · Zero-Reference Deep Curve Estimation for Low-Light Image Enhancement (CVPR 2020 · 2,412 citations)
- 2021 · Nighttime Visibility Enhancement by Increasing the Dynamic Range and Suppression of Light Effects (CVPR 2021 · 79 citations)
- 2022 · Toward Fast, Flexible, and Robust Low-Light Image Enhancement (CVPR 2022 · 1,068 citations)
- 2023 · Ultra-High-Definition Low-Light Image Enhancement: A Benchmark and Transformer-Based Method (AAAI 2023 · 523 citations)
- 2024 · Zero-Reference Low-Light Enhancement via Physical Quadruple Priors (CVPR 2024 · 135 citations)
- 2025 · HVI: A New Color Space for Low-light Image Enhancement (CVPR 2025 · 212 citations)
- 2026 · SPJFNet: Self-Mining Prior-Guided Joint Frequency Enhancement for Ultra-Efficient Dark Image Restoration (AAAI 2026 · 4 citations)
Earliest papers here
- Fast Haze Removal for Nighttime Image Using Maximum Reflectance Prior (CVPR 2017 · 206 citations)
- Learning Dynamic Guidance for Depth Image Enhancement (CVPR 2017 · 103 citations)
- Learning to See in the Dark (CVPR 2018 · 1,403 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
LOL · PSNR (higher is better)
11 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 21.33 | URetinex-Net: Retinex-Based Deep Unfolding Network for Low-Light Image Enhancement (CVPR 2022 · 797 citations) |
| 2023 | 26.8 | Learning Semantic-Aware Knowledge Guidance for Low-Light Image Enhancement (CVPR 2023 · 200 citations) |
| 2024 | 27.12 | Multi-Domain Multi-Scale Diffusion Model for Low-Light Image Enhancement (AAAI 2024 · 21 citations) |
| 2025 | 27.51 | Low-Light Image Enhancement via Generative Perceptual Priors (AAAI 2025 · 23 citations) |
LOL · SSIM (higher is better)
8 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 0.835 | URetinex-Net: Retinex-Based Deep Unfolding Network for Low-Light Image Enhancement (CVPR 2022 · 797 citations) |
| 2023 | 0.928 | You Do Not Need Additional Priors or Regularizers in Retinex-Based Low-Light Image Enhancement (CVPR 2023 · 84 citations) |
LOLv1 · PSNR (higher is better)
8 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 24.61 | SNR-Aware Low-Light Image Enhancement (CVPR 2022 · 676 citations) |
| 2023 | 27.84 | Global Structure-Aware Diffusion Process for Low-light Image Enhancement (NeurIPS 2023 · 78 citations) |
| 2025 | 28.2 | HVI: A New Color Space for Low-light Image Enhancement (CVPR 2025 · 212 citations) |
DICM · NIQE (lower is better)
7 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 3.7 | Deep Color Consistent Network for Low-Light Image Enhancement (CVPR 2022 · 155 citations) |
| 2023 | 3.28 | Global Structure-Aware Diffusion Process for Low-light Image Enhancement (NeurIPS 2023 · 78 citations) |
| 2024 | 3.23 | Specularity Factorization for Low-Light Enhancement (CVPR 2024 · 16 citations) |
LOL · LPIPS (lower is better)
7 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 0.107 | You Do Not Need Additional Priors or Regularizers in Retinex-Based Low-Light Image Enhancement (CVPR 2023 · 84 citations) |
| 2024 | 0.078 | Multi-Domain Multi-Scale Diffusion Model for Low-Light Image Enhancement (AAAI 2024 · 21 citations) |
LOLv1 · SSIM (higher is better)
6 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 0.877 | Global Structure-Aware Diffusion Process for Low-light Image Enhancement (NeurIPS 2023 · 78 citations) |
| 2025 | 0.889 | HVI: A New Color Space for Low-light Image Enhancement (CVPR 2025 · 212 citations) |
What has been tried
60 papers here report 593 measured changes (a component added, removed or swapped, compared with a baseline): helped 80%, no clear effect 4%, hurt 15%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| replace sRGB with HVI color space | LOLv2-Real · PSNR | 20.062 -> 24.111 | Helped | HVI: A New Color Space for Low-light Image Enhancement (CVPR 2025 · 212 citations) |
| add H to the physical quadruple prior | LOL · PSNR | 17.60 -> 20.25 | Helped | Zero-Reference Low-Light Enhancement via Physical Quadruple Priors (CVPR 2024 · 135 citations) |
| Add the reflection-illumination concatenation operation | SIDD-Small · PSNR | 19.4880 -> 23.2349 | Helped | ZERO-IG: Zero-Shot Illumination-Guided Joint Denoising and Adaptive Enhancement for Low-Light Images (CVPR 2024 · 94 citations) |
| add spatial-frequency alternating optimization | RealBlur night · NIQE | 3.62 -> 3.25 | Helped | Fourier Priors-Guided Diffusion for Zero-Shot Joint Low-Light Enhancement and Deblurring (CVPR 2024 · 83 citations) |
| add LiDAR and Distribution Reward Modeling (LDRM) | nuScenes nighttime validation · AP | 0.151 -> 0.176 | Helped | Light the Night: A Multi-Condition Diffusion Framework for Unpaired Low-Light Enhancement in Autonomous Driving (CVPR 2024 · 79 citations) |
| add events to the Base model | SDE-in · PSNR | +0.23 dB | Helped | Towards Robust Event-guided Low-Light Image Enhancement: A Large-Scale Real-World Event-Image Dataset and Novel Approach (CVPR 2024 · 45 citations) |
| replace color-space deformable convolution with standard convolution | LCDP test set · PSNR | 23.443 -> 23.627 | Helped | Color Shift Estimation-and-Correction for Image Enhancement (CVPR 2024 · 40 citations) |
| Add exposure attribute guidance | LOLv1 · exposure control | Helped | AGLLDiff: Guiding Diffusion Models Towards Unsupervised Training-free Real-world Low-light Image Enhancement (AAAI 2025 · 24 citations) | |
| add the local perceptual prior using LPP-Attn | LOL · FID | 49.83 -> 36.73 | Helped | Low-Light Image Enhancement via Generative Perceptual Priors (AAAI 2025 · 23 citations) |
| Remove rectified flow (direct distillation) | CelebA-Test · FID | 19.81 -> 49.74 | Hurt | FlowIE: Efficient Image Enhancement via Rectified Flow (CVPR 2024 · 23 citations) |
| add the spatial branch | LOL · PSNR | 26.36 -> 27.12 | Helped | Multi-Domain Multi-Scale Diffusion Model for Low-Light Image Enhancement (AAAI 2024 · 21 citations) |
| Add exposure loss | LOLv1 · PSNRy | 8.12 -> 22.17 | Helped | Specularity Factorization for Low-Light Enhancement (CVPR 2024 · 16 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| SNR-Aware | LOLv1 | PSNR | 24.61 – 26.72 | 3 papers, e.g. AAAI 2025 and NeurIPS 2023 |
| Pydiff | LOLv2-Syn | PSNR | 17.33 – 29.27 | 2 papers, e.g. AAAI 2024 and AAAI 2025 |
| GSAD | LOL-v1 | LPIPS | 0.092 – 0.104 | 2 papers, e.g. CVPR 2025 and AAAI 2025 |
| DiffLL | LOL | PSNR | 26.19 – 28.54 | 2 papers, e.g. AAAI 2025 and CVPR 2024 |
| Pydiff | LOLv2-Real | PSNR | 28.77 – 31.11 | 2 papers, e.g. AAAI 2025 and AAAI 2024 |
| URetinexNet | LOL | PSNR | 19.84 – 21.33 | 2 papers, e.g. CVPR 2023 and CVPR 2023 |
What reviewers will expect
Datasets and baselines named in the 13 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- LOLv1 · 6 of 13 papers
- LIME · 4 of 13 papers
- NPE · 4 of 13 papers
- DICM · 4 of 13 papers
- SICE · 3 of 13 papers
- MEF · 3 of 13 papers
- VV · 3 of 13 papers
- ImageNet · 3 of 13 papers
Baselines
- Retinexformer · 7 of 13 papers
- PairLIE · 7 of 13 papers
- EnlightenGAN · 6 of 13 papers
- SCI · 6 of 13 papers
- NeRCo · 6 of 13 papers
- Zero-DCE · 6 of 13 papers
- RetinexNet · 5 of 13 papers
- RUAS · 5 of 13 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 47 papers; from OpenAlex and the papers' affiliations).
- Nanyang Technological University · 4 papers
- Peking University · 3 papers
- Xiamen University · 3 papers
- Hong Kong University of Science and Technology · 3 papers
- University of Science and Technology of China · 2 papers
- Jiangxi University of Finance and Economics · 2 papers
- Beijing University of Posts and Telecommunications · 2 papers
- Nanjing University · 2 papers
- South China University of Technology · 2 papers
- Dalian University of Technology · 2 papers
Use it
Download BibTeX of 38 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Dehazing and deraining (118 papers)
- Inverse imaging with learned priors (109 papers)
- Image harmonization and color transformation (102 papers)
- Reflection removal (95 papers)
- Pansharpening and spectral fusion (84 papers)
Elsewhere in the topic
- Efficient restoration architectures (113 papers)
- Image deblurring (73 papers)
- Learned image compression (69 papers)
- Priors for restoration networks (66 papers)
- Diffusion-based universal restoration (61 papers)
- Self-supervised image denoising (60 papers)
- Raw and low-light noise modeling (53 papers)
- Diffusion priors for denoising and deblurring (41 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).