Research map / Image generation / Image restoration
Image deblurring
Remove blur with blind, unsupervised, and kernel-aware deblurring networks.
73 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Image restoration, keyword cluster “image denoising · blur · noise”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 6 tracked benchmarks.
26 papers in 2023–24 and 10 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 8.4% to 2.3%.
Open questions the data points to
- GoPro (PSNR): best went from 35.13 (up to 2024) to 35.5 in 2025 — Diffusion-based Event Generation for High-Quality Image Deblurring (CVPR 2025 · 10 citations)
- Papers report UFPNet on RealBlur-R (PSNR) anywhere from 36.25 to 40.61: comparisons against it depend on whose number you take.
- Papers report UFPNet on RealBlur-J (PSNR) anywhere from 29.87 to 33.35: comparisons against it depend on whose number you take.
What looks settled
- RealBlur-R (PSNR): best up to 2024 41.23, best in 2025–26 41 from 3 papers, at most 0.2% better, if at all.
- RealBlur-J (PSNR): best up to 2024 33.88, best in 2025–26 33.87 from 3 papers, at most 0.2% better, if at all.
- HIDE (PSNR): best up to 2024 31.94, best in 2025–26 31.92 from 3 papers, at most 0.2% better, if at all.
No longer reported by papers here: GoPro test (PSNR, last in 2024).
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 · Blind Image Deblurring Using Dark Channel Prior (CVPR 2016 · 825 citations)
- 2017 · From Motion Blur to Motion Flow: A Deep Learning Solution for Removing Heterogeneous Motion Blur (CVPR 2017 · 466 citations)
- 2018 · Scale-Recurrent Network for Deep Image Deblurring (CVPR 2018 · 1,365 citations)
- 2019 · Deep Stacked Hierarchical Multi-Patch Network for Image Deblurring (CVPR 2019 · 659 citations)
- 2020 · Deblurring by Realistic Blurring (CVPR 2020 · 450 citations)
- 2021 · Iterative Filter Adaptive Network for Single Image Defocus Deblurring (CVPR 2021 · 179 citations)
- 2022 · Deblurring via Stochastic Refinement (CVPR 2022 · 260 citations)
- 2023 · Efficient Frequency Domain-Based Transformers for High-Quality Image Deblurring (CVPR 2023 · 387 citations)
- 2024 · Efficient Multi-scale Network with Learnable Discrete Wavelet Transform for Blind Motion Deblurring (CVPR 2024 · 78 citations)
- 2025 · Multi-Focus Image Fusion via Explicit Defocus Blur Modelling (AAAI 2025 · 10 citations)
- 2026 · BluRef: Unsupervised Image Deblurring with Dense-Matching References (CVPR 2026 · 0 citations)
Earliest papers here
- Blind Image Deblurring Using Dark Channel Prior (CVPR 2016 · 825 citations)
- A Comparative Study for Single Image Blind Deblurring (CVPR 2016 · 464 citations)
- Robust Kernel Estimation With Outliers Handling for Image Deblurring (CVPR 2016 · 113 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.
GoPro · PSNR (higher is better)
12 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 35.13 | Explore Image Deblurring via Encoded Blur Kernel Space (CVPR 2021 · 76 citations) |
| 2025 | 35.5 | Diffusion-based Event Generation for High-Quality Image Deblurring (CVPR 2025 · 10 citations) |
RealBlur-R · PSNR (higher is better)
10 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 40.11 | Efficient Frequency Domain-Based Transformers for High-Quality Image Deblurring (CVPR 2023 · 387 citations) |
| 2024 | 41.23 | Motion-adaptive Separable Collaborative Filters for Blind Motion Deblurring (CVPR 2024 · 26 citations) |
RealBlur-J · PSNR (higher is better)
9 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 32.62 | Efficient Frequency Domain-Based Transformers for High-Quality Image Deblurring (CVPR 2023 · 387 citations) |
| 2024 | 33.88 | Motion-adaptive Separable Collaborative Filters for Blind Motion Deblurring (CVPR 2024 · 26 citations) |
HIDE · PSNR (higher is better)
8 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 31.62 | Efficient Frequency Domain-Based Transformers for High-Quality Image Deblurring (CVPR 2023 · 387 citations) |
| 2024 | 31.94 | ID-Blau: Image Deblurring by Implicit Diffusion-based reBLurring AUgmentation (CVPR 2024 · 22 citations) |
GoPro test · PSNR (higher is better)
7 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 30.1 | Scale-Recurrent Network for Deep Image Deblurring (CVPR 2018 · 1,365 citations) |
| 2019 | 31.5 | Deep Stacked Hierarchical Multi-Patch Network for Image Deblurring (CVPR 2019 · 659 citations) |
| 2023 | 34.06 | Self-Supervised Non-Uniform Kernel Estimation With Flow-Based Motion Prior for Blind Image Deblurring (CVPR 2023 · 61 citations) |
| 2024 | 34.38 | Real-World Efficient Blind Motion Deblurring via Blur Pixel Discretization (CVPR 2024 · 11 citations) |
GoPro · SSIM (higher is better)
6 papers, 2022–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 0.963 | Deblurring via Stochastic Refinement (CVPR 2022 · 260 citations) |
| 2023 | 0.969 | Efficient Frequency Domain-Based Transformers for High-Quality Image Deblurring (CVPR 2023 · 387 citations) |
| 2025 | 0.974 | Diffusion-based Event Generation for High-Quality Image Deblurring (CVPR 2025 · 10 citations) |
What has been tried
71 papers here report 673 measured changes (a component added, removed or swapped, compared with a baseline): helped 81%, no clear effect 4%, hurt 14%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use SIMO instead of SISO | RealBlur-J ablation · PSNR | 32.29 -> 32.37 | Helped | Efficient Multi-scale Network with Learnable Discrete Wavelet Transform for Blind Motion Deblurring (CVPR 2024 · 78 citations) |
| Add FourierBlock instead of NAFBlock | GoPro test · PSNR | 34.15 -> 34.51 | Helped | AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring (CVPR 2024 · 33 citations) |
| add motion-guided alignment | GoPro · PSNR | 32.40 -> 32.51 | Helped | Motion-adaptive Separable Collaborative Filters for Blind Motion Deblurring (CVPR 2024 · 26 citations) |
| Use a 6:4 rather than 5:5 unknown-blurry-to-sharp data ratio | GoPro–RB2V Street · PSNR | 26.02 -> 26.98 | Helped | Blur2Blur: Blur Conversion for Unsupervised Image Deblurring on Unknown Domains (CVPR 2024 · 25 citations) |
| Rotate blur orientations by 30°, 60°, 90°, 120°, or a random angle | GoPro · PSNR | 31.22 -> 31.99; 31.22 -> 32.00; 31.22 -> 31.97; 31.22 -> 31.95; 31.22 -> 31.96 | Helped | ID-Blau: Image Deblurring by Implicit Diffusion-based reBLurring AUgmentation (CVPR 2024 · 22 citations) |
| Add self-enhancement | GoPro · PSNR | 27.68 -> 29.06 | Helped | Unsupervised Blind Image Deblurring Based on Self-Enhancement (CVPR 2024 · 17 citations) |
| use blur-map kernel priors in the BPA block instead of concatenating the blur map and using standard self-attention | DPD-blur · PSNR | 26.80 -> 26.91 | Helped | LDP: Language-driven Dual-Pixel Image Defocus Deblurring Network (CVPR 2024 · 16 citations) |
| Set λ to 1 rather than 3 | GaoFen2 reduced-resolution · PSNR | 35.71 -> 32.08 | Hurt | Unsupervised Pan-Sharpening via Mutually Guided Detail Restoration (AAAI 2024 · 12 citations) |
| Add the full blur-segmentation-map, latent-sharp-loss, and logarithmic-Fourier components | RealBlur-J test · PSNR | 31.99 -> 32.53 | Helped | Real-World Efficient Blind Motion Deblurring via Blur Pixel Discretization (CVPR 2024 · 11 citations) |
| Use a coarse-to-fine multiscale DBE instead of one DBE scale | Lytro · ARank | 2 -> 1.4 | Helped | Multi-Focus Image Fusion via Explicit Defocus Blur Modelling (AAAI 2025 · 10 citations) |
| Generate dense motion-prior features rather than raw events | GoPro · PSNR | 32.33 -> 34.62 | Helped | Diffusion-based Event Generation for High-Quality Image Deblurring (CVPR 2025 · 10 citations) |
| Add IKPM to SRT | DPDD · PSNR | 26.03 -> 26.12 | Helped | Prior and Prediction Inverse Kernel Transformer for Single Image Defocus Deblurring (AAAI 2024 · 8 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 |
|---|---|---|---|---|
| UFPNet | RealBlur-R | PSNR | 36.25 – 40.61 | 2 papers, e.g. CVPR 2024 and CVPR 2024 |
| UFPNet | RealBlur-J | PSNR | 29.87 – 33.35 | 2 papers, e.g. CVPR 2024 and CVPR 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
- GoPro · 4 of 9 papers
- DPDD · 2 of 9 papers
- HIDE · 2 of 9 papers
- RealBlur-J · 2 of 9 papers
- RealBlur-R · 2 of 9 papers
Baselines
- FFTformer · 5 of 9 papers
- Restormer · 4 of 9 papers
- Stripformer · 4 of 9 papers
- Uformer · 3 of 9 papers
- NAFNet · 3 of 9 papers
- DPDNet · 2 of 9 papers
- DRBNet · 2 of 9 papers
- MPRNet · 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 36 papers; from OpenAlex and the papers' affiliations).
- Peking University · 4 papers
- National University of Singapore · 4 papers
- Harbin Institute of Technology · 2 papers
- University of Electronic Science and Technology of China · 2 papers
- South China University of Technology · 2 papers
- Tsinghua University · 2 papers
- National Tsing Hua University · 2 papers
- National Chengchi University · 2 papers
- National Yang Ming Chiao Tung University · 2 papers
- Australian National 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
- Self-supervised image denoising (60 papers)
- Raw and low-light noise modeling (53 papers)
- Diffusion priors for denoising and deblurring (41 papers)
Elsewhere in the topic
- Dehazing and deraining (118 papers)
- Efficient restoration architectures (113 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)
- 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).