Research map / Image generation / Image restoration
Real-world restoration with unaligned data
Restore images in dynamic scenes, burst, and under-display cameras using pseudo-supervision.
40 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Image restoration, keyword cluster “image restoration · degradation · degraded”
State History Tried Baselines 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.
12 papers in 2023–24 and 16 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 3.9% to 3.7%.
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 · Exploiting Spectral-Spatial Correlation for Coded Hyperspectral Image Restoration (CVPR 2016 · 108 citations)
- 2019 · Attention-Based Adaptive Selection of Operations for Image Restoration in the Presence of Unknown Combined Distortions (CVPR 2019 · 104 citations)
- 2020 · Enhanced Blind Face Restoration With Multi-Exemplar Images and Adaptive Spatial Feature Fusion (CVPR 2020 · 127 citations)
- 2021 · Image Restoration for Under-Display Camera (CVPR 2021 · 69 citations)
- 2022 · Burst Image Restoration and Enhancement (CVPR 2022 · 102 citations)
- 2023 · Burstormer: Burst Image Restoration and Enhancement Transformer (CVPR 2023 · 60 citations)
- 2024 · VQCNIR: Clearer Night Image Restoration with Vector-Quantized Codebook (AAAI 2024 · 30 citations)
- 2025 · VolFormer: Explore More Comprehensive Cube Interaction for Hyperspectral Image Restoration and Beyond (CVPR 2025 · 12 citations)
- 2026 · Clear Nights Ahead: Towards Multi-Weather Nighttime Image Restoration (AAAI 2026 · 1 citation)
Earliest papers here
- Exploiting Spectral-Spatial Correlation for Coded Hyperspectral Image Restoration (CVPR 2016 · 108 citations)
- Needle-Match: Reliable Patch Matching Under High Uncertainty (CVPR 2016 · 26 citations)
- Attention-Based Adaptive Selection of Operations for Image Restoration in the Presence of Unknown Combined Distortions (CVPR 2019 · 104 citations)
What has been tried
31 papers here report 336 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 3%, hurt 21%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add decoder D to the VQGAN codebook prior | LOL-Blur · PSNR | 10.79 -> 26.58 | Helped | VQCNIR: Clearer Night Image Restoration with Vector-Quantized Codebook (AAAI 2024 · 30 citations) |
| Increase diagonal width λ from 1 to 5 | Set5 ×4 super-resolution · PSNR | Helped | Look-Up Table Compression for Efficient Image Restoration (CVPR 2024 · 14 citations) | |
| add volumetric self-attention instead of spatial self-attention | CAVE ×4 · MPSNR | 39.597 -> 40.205 | Helped | VolFormer: Explore More Comprehensive Cube Interaction for Hyperspectral Image Restoration and Beyond (CVPR 2025 · 12 citations) |
| Add fixed-mask frequency mining to baseline | SOTS dehazing ablation · PSNR | 28.21 -> 29.79 | Helped | AdaIR: Adaptive All-in-One Image Restoration via Frequency Mining and Modulation (ICLR 2025 · 9 citations) |
| Remove two-stage training | WV3 · SAM | 2.917±0.560 -> 3.573±0.682 | Hurt | PanAdapter: Two-Stage Fine-Tuning with Spatial-Spectral Priors Injecting for Pansharpening (AAAI 2025 · 7 citations) |
| Add pre-correction to Stripformer | LAP dataset · PSNR (dB) | 37.74 -> 38.72 | Helped | Deep Linear Array Pushbroom Image Restoration: A Degradation Pipeline and Jitter-Aware Restoration Network (AAAI 2024 · 7 citations) |
| Remove the pre-trained encoder and decoder (exp1) | MDLP · PSNR | 14.396 -> 13.757 | Hurt | LP-Diff: Towards Improved Restoration of Real-World Degraded License Plate (CVPR 2025 · 5 citations) |
| Add color-corrected image conditioning (CIC) to DDPM | DESCAN-18K test · PSNR (dB) | 22.72 -> 23.18 | Helped | Descanning: From Scanned to the Original Images with a Color Correction Diffusion Model (AAAI 2024 · 4 citations) |
| Increase the number of training scenes from 10 to 240 | SIDL test set (15 scenes) · PSNR | 20.18 / 0.75 -> 23.73 / 0.86 | Helped | SIDL: A Real-World Dataset for Restoring Smartphone Images with Dirty Lenses (AAAI 2025 · 2 citations) |
| add dynamic recalibration | OpenSRH unpaired images · FID | 34.62 -> 32.02 | Helped | Step-Calibrated Diffusion for Biomedical Optical Image Restoration (AAAI 2025 · 1 citation) |
| Add an auxiliary LQ encoder alongside the HQ encoder | CelebA-Test · NIQE | 4.297 vs 4.394 | Helped | Dual Associated Encoder for Face Restoration (ICLR 2024 · 1 citation) |
| use DINOv2 or CLIP features instead of FairFace features to compute GPI-KID | synthetic face groups · GPI-KID group disparity | Mixed | Perceptual Fairness in Image Restoration (NeurIPS 2024 · 1 citation) |
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 |
|---|---|---|---|---|
| CodeFormer | CelebA-Test | LPIPS | 0.343 – 0.365 | 2 papers, e.g. ICLR 2025 and ICLR 2024 |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- DIV2K · 3 of 9 papers
- WED · 2 of 9 papers
- FFHQ · 2 of 9 papers
- CelebA-HQ · 2 of 9 papers
- WebPhoto-Test · 2 of 9 papers
Baselines
- AirNet · 2 of 9 papers
- NAFNet · 2 of 9 papers
- Restormer · 2 of 9 papers
- GFPGAN · 2 of 9 papers
- CodeFormer · 2 of 9 papers
- RestoreFormer · 2 of 9 papers
- DifFace · 2 of 9 papers
- DiffBIR · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 28 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 2 papers
- Technion · 2 papers
Use it
Download BibTeX of 34 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
- Efficient restoration architectures (113 papers)
- Priors for restoration networks (66 papers)
- Diffusion-based universal restoration (61 papers)
Elsewhere in the topic
- 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)
- Image deblurring (73 papers)
- Low-light image enhancement (72 papers)
- Learned image compression (69 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).