Research map / Image generation / Image restoration
Reflection removal
Separate reflections from transmission using misaligned data, raw inputs, and flash cues.
95 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Image restoration, keyword cluster “removal · enhancement · color”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
20 papers in 2023–24 and 36 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 6.4% to 8.3%.
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
- 2016 · Geospatial Correspondences for Multimodal Registration (CVPR 2016 · 40 citations)
- 2017 · Single Image Reflection Suppression (CVPR 2017 · 156 citations)
- 2018 · A Revised Underwater Image Formation Model (CVPR 2018 · 460 citations)
- 2019 · Sea-Thru: A Method for Removing Water From Underwater Images (CVPR 2019 · 628 citations)
- 2020 · Contextual Residual Aggregation for Ultra High-Resolution Image Inpainting (CVPR 2020 · 388 citations)
- 2021 · Learning Multi-Scale Photo Exposure Correction (CVPR 2021 · 229 citations)
- 2022 · Learning Invisible Markers for Hidden Codes in Offline-to-Online Photography (CVPR 2022 · 38 citations)
- 2023 · Nighttime Smartphone Reflective Flare Removal Using Optical Center Symmetry Prior (CVPR 2023 · 33 citations)
- 2024 · Devignet: High-Resolution Vignetting Removal via a Dual Aggregated Fusion Transformer with Adaptive Channel Expansion (AAAI 2024 · 44 citations)
- 2025 · UltraFusion: Ultra High Dynamic Imaging using Exposure Fusion (CVPR 2025 · 23 citations)
- 2026 · Gradient as Conditions: Rethinking HOG for All-in-one Image Restoration (AAAI 2026 · 2 citations)
Earliest papers here
- Geospatial Correspondences for Multimodal Registration (CVPR 2016 · 40 citations)
- Single Image Reflection Suppression (CVPR 2017 · 156 citations)
- Reflection Removal Using Low-Rank Matrix Completion (CVPR 2017 · 80 citations)
What has been tried
77 papers here report 739 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 3%, hurt 17%, 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 |
|---|---|---|---|---|
| Add ACEM for high-frequency detail reconstruction | VigSet 512×512 · PSNR | 19.66 -> 22.96 | Helped | Devignet: High-Resolution Vignetting Removal via a Dual Aggregated Fusion Transformer with Adaptive Channel Expansion (AAAI 2024 · 44 citations) |
| Add RRW extra training data to RDNet + RRNet | Nature · PSNR | 25.37 -> 25.96 | Helped | Revisiting Single Image Reflection Removal In the Wild (CVPR 2024 · 33 citations) |
| Fine-tune the entire SAM model on degraded images | BDD-100k+LIS, point prompts · IoU | 0.3056 -> 0.1871 | Hurt | RobustSAM: Segment Anything Robustly on Degraded Images (CVPR 2024 · 33 citations) |
| add background-image lighting conditioning | Light stage test set · PSNR | 18.768 -> 19.748 | Helped | Relightful Harmonization: Lighting-aware Portrait Background Replacement (CVPR 2024 · 28 citations) |
| Use only low-resolution pixel-wise transformation | LCDP · PSNR | 20.08 -> 23.89 | Hurt | Real-Time Exposure Correction via Collaborative Transformations and Adaptive Sampling (CVPR 2024 · 24 citations) |
| Add the alignment strategy (pre-alignment and its training data synthesis pipeline) | RealHDRV · TMQI | 0.7427 -> 0.8925 | Helped | UltraFusion: Ultra High Dynamic Imaging using Exposure Fusion (CVPR 2025 · 23 citations) |
| add Structural Loss | UIEB test · SSIM | 0.883 -> 0.894 | Helped | Underwater Organism Color Fine-Tuning via Decomposition and Guidance (AAAI 2024 · 23 citations) |
| Use spatially varying deconvolution rather than a single-kernel deconvolution | PhlatCam and DiffuserCam · reconstruction fidelity | Helped | PhoCoLens: Photorealistic and Consistent Reconstruction in Lensless Imaging (NeurIPS 2024 · 18 citations) | |
| Add phase-based loss | PolaRGB test · PSNR | 22.41 -> 22.44 | Helped | PolarFree: Polarization-based Reflection-Free Imaging (CVPR 2025 · 12 citations) |
| Add VAE decoder fine-tuning | SI-HDR highlights · HDR-VDP3 | 4.67 ± 0.87 -> 6.16 ± 0.97 | Helped | LEDiff: Latent Exposure Diffusion for HDR Generation (CVPR 2025 · 11 citations) |
| Separate stage-1 prediction of the background component from watermark-free image prediction | HWVOC · PSNR | 38.74 -> 39.52 | Helped | Removing Interference and Recovering Content Imaginatively for Visible Watermark Removal (AAAI 2024 · 10 citations) |
| add frequency-domain processing | Our Dataset (HAADF mode) · PSNR | 34.61 -> 38.48 | Helped | Noise Calibration and Spatial-Frequency Interactive Network for STEM Image Enhancement (CVPR 2025 · 8 citations) |
What reviewers will expect
Datasets and baselines named in the 24 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- SICE · 2 of 24 papers
- COCO · 2 of 24 papers
Baselines
- DSRNet · 3 of 24 papers
- HDR-Transformer · 3 of 24 papers
- Zhang et al. · 2 of 24 papers
- IBCLN · 2 of 24 papers
- StegaStamp · 2 of 24 papers
- BIPNet · 2 of 24 papers
- Kim et al. · 2 of 24 papers
- SCTNet · 2 of 24 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 56 papers; from OpenAlex and the papers' affiliations).
- The Chinese University of Hong Kong · 4 papers
- Peking University · 4 papers
- University of Science and Technology of China · 4 papers
- Shanghai AI Laboratory · 3 papers
- The Hong Kong Polytechnic University · 3 papers
- Xiamen University · 2 papers
- Shanghai Jiao Tong University · 2 papers
- Chinese Academy of Sciences · 2 papers
- Chongqing University of Posts and Telecommunications · 2 papers
- Xidian University · 2 papers
Use it
Download BibTeX of 40 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)
- Pansharpening and spectral fusion (84 papers)
- Low-light image enhancement (72 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).