Research map / Image generation / Image restoration
Efficient restoration architectures
Design efficient, progressive, and frequency-based networks for general image restoration.
113 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Image restoration, keyword cluster “image restoration · degradation · degraded”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 4 of 6 tracked benchmarks.
30 papers in 2023–24 and 65 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 9.6% to 14.9%.
Open questions the data points to
- GoPro (PSNR): best went from 34.05 (up to 2024) to 34.53 in 2025 — Visual-Instructed Degradation Diffusion for All-in-One Image Restoration (CVPR 2025 · 7 citations)
- SIDD (PSNR): best went from 40 (up to 2024) to 42.26 in 2025 — Reversing Flow for Image Restoration (CVPR 2025 · 7 citations)
- RainDrop (PSNR): best went from 32.89 (up to 2024) to 33.81 in 2025 — Visual-Instructed Degradation Diffusion for All-in-One Image Restoration (CVPR 2025 · 7 citations)
- LOL (PSNR): best went from 23.57 (up to 2024) to 25.48 in 2025 — Intra and Inter Parser-Prompted Transformers for Effective Image Restoration (AAAI 2025 · 7 citations)
No longer reported by papers here: HIDE (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
- 2018 · The Perception-Distortion Tradeoff (CVPR 2018 · 994 citations)
- 2019 · Dynamically Unfolding Recurrent Restorer: A Moving Endpoint Control Method for Image Restoration (ICLR 2019 · 21 citations)
- 2020 · All in One Bad Weather Removal Using Architectural Search (CVPR 2020 · 360 citations)
- 2021 · Multi-Stage Progressive Image Restoration (CVPR 2021 · 2,086 citations)
- 2022 · Unsupervised Underwater Image Restoration: From a Homology Perspective (AAAI 2022 · 155 citations)
- 2023 · Efficient and Explicit Modelling of Image Hierarchies for Image Restoration (CVPR 2023 · 292 citations)
- 2024 · Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild (CVPR 2024 · 139 citations)
- 2025 · MambaIRv2: Attentive State Space Restoration (CVPR 2025 · 140 citations)
- 2026 · Pansharpening for Thin-Cloud Contaminated Remote Sensing Images: A Unified Framework and Benchmark Dataset (AAAI 2026 · 14 citations)
Earliest papers here
- The Perception-Distortion Tradeoff (CVPR 2018 · 994 citations)
- Noise2Noise: Learning Image Restoration without Clean Data (ICML 2018 · 982 citations)
- Multi-Scale Weighted Nuclear Norm Image Restoration (CVPR 2018 · 90 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)
9 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 32.66 | Multi-Stage Progressive Image Restoration (CVPR 2021 · 2,086 citations) |
| 2022 | 33.48 | GIQE: Generic Image Quality Enhancement via Nth Order Iterative Degradation (CVPR 2022 · 10 citations) |
| 2023 | 33.93 | Efficient and Explicit Modelling of Image Hierarchies for Image Restoration (CVPR 2023 · 292 citations) |
| 2024 | 34.05 | CoDe: An Explicit Content Decoupling Framework for Image Restoration (CVPR 2024 · 6 citations) |
| 2025 | 34.53 | Visual-Instructed Degradation Diffusion for All-in-One Image Restoration (CVPR 2025 · 7 citations) |
SIDD · PSNR (higher is better)
6 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 39.71 | Multi-Stage Progressive Image Restoration (CVPR 2021 · 2,086 citations) |
| 2023 | 40 | Random Shuffle Transformer for Image Restoration (ICML 2023) |
| 2025 | 42.26 | Reversing Flow for Image Restoration (CVPR 2025 · 7 citations) |
Rain100L · PSNR (higher is better)
5 papers, 2021–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 36.4 | Multi-Stage Progressive Image Restoration (CVPR 2021 · 2,086 citations) |
| 2023 | 37.47 | Fourmer: An Efficient Global Modeling Paradigm for Image Restoration (ICML 2023) |
| 2024 | 39.22 | EnsIR: An Ensemble Algorithm for Image Restoration via Gaussian Mixture Models (NeurIPS 2024 · 2 citations) |
RainDrop · PSNR (higher is better)
5 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 25.18 | GIQE: Generic Image Quality Enhancement via Nth Order Iterative Degradation (CVPR 2022 · 10 citations) |
| 2023 | 31.31 | Learning Weather-General and Weather-Specific Features for Image Restoration Under Multiple Adverse Weather Conditions (CVPR 2023 · 167 citations) |
| 2024 | 32.89 | IR-CM: The Fast and General-purpose Image Restoration Method Based on Consistency Model (NeurIPS 2024 · 0 citations) |
| 2025 | 33.81 | Visual-Instructed Degradation Diffusion for All-in-One Image Restoration (CVPR 2025 · 7 citations) |
HIDE · PSNR (higher is better)
4 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 30.96 | Multi-Stage Progressive Image Restoration (CVPR 2021 · 2,086 citations) |
| 2023 | 31.65 | Efficient and Explicit Modelling of Image Hierarchies for Image Restoration (CVPR 2023 · 292 citations) |
| 2024 | 31.76 | CoDe: An Explicit Content Decoupling Framework for Image Restoration (CVPR 2024 · 6 citations) |
LOL · PSNR (higher is better)
4 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 17.5 | Zero-Shot Single Image Restoration Through Controlled Perturbation of Koschmieder's Model (CVPR 2021) |
| 2023 | 23.57 | Fourmer: An Efficient Global Modeling Paradigm for Image Restoration (ICML 2023) |
| 2025 | 25.48 | Intra and Inter Parser-Prompted Transformers for Effective Image Restoration (AAAI 2025 · 7 citations) |
What has been tried
74 papers here report 821 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 3%, hurt 21%, 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 ASE to window MHSA | Urban100, 2× SR · PSNR | 32.89 -> 32.94 | Helped | MambaIRv2: Attentive State Space Restoration (CVPR 2025 · 140 citations) |
| Replace ZeroSFT with zero convolution | Real-world LQ images · PSNR | 20.66 -> 19.47 | Hurt | Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild (CVPR 2024 · 139 citations) |
| add SDT to DiffUIR | five restoration tasks · PSNR | 30.34 -> 31.03; 22.99 -> 25.12; 32.34 -> 32.65; 32.47 -> 32.94; 28.50 -> 29.17 | Helped | Selective Hourglass Mapping for Universal Image Restoration Based on Diffusion Model (CVPR 2024 · 74 citations) |
| Add the textual branch | Deraining, dehazing, raindrop removal average · PSNR | 16.49 -> 19.19 | Helped | Multimodal Prompt Perceiver: Empower Adaptiveness Generalizability and Fidelity for All-in-One Image Restoration (CVPR 2024 · 68 citations) |
| add DualCMT to CMTA and CMTN | UHD-LL · PSNR | 25.431 -> 27.113 | Helped | Correlation Matching Transformation Transformers for UHD Image Restoration (AAAI 2024 · 61 citations) |
| Replace the proposed scanning with Z-shaped scanning | Urban100 ×2 lightweight SR · PSNR | 32.97 -> 32.94 | Hurt | MaIR: A Locality- and Continuity-Preserving Mamba for Image Restoration (CVPR 2025 · 59 citations) |
| Add DTSPrompt to the unified Restormer baseline | DIR300 · MS-SSIM | 0.6030 -> 0.6264 | Helped | DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks (CVPR 2024 · 35 citations) |
| Add cascaded fIR2 | Deraining ablation setting · PSNR | Hurt | Distilling Semantic Priors from SAM to Efficient Image Restoration Models (CVPR 2024 · 28 citations) | |
| set textual representation word count N to 5 | GoPro motion deblurring · PSNR | 30.16 -> 31.13 | Helped | Improving Image Restoration through Removing Degradations in Textual Representations (CVPR 2024 · 28 citations) |
| add channel attention only | Three-task average · PSNR (dB) | 31.98 -> 32.34 | Helped | Degradation-Aware Feature Perturbation for All-in-One Image Restoration (CVPR 2025 · 19 citations) |
| Add LAN | LOL-v2-real · PSNR | 21.33 -> 22.71 | Helped | URWKV: Unified RWKV Model with Multi-state Perspective for Low-light Image Restoration (CVPR 2025 · 18 citations) |
| Replace patch-based degradation learning with DSL | GFIQA-20k · PLCC | 0.9695 -> 0.9721 | Helped | DSL-FIQA: Assessing Facial Image Quality via Dual-Set Degradation Learning and Landmark-Guided Transformer (CVPR 2024 · 17 citations) |
What reviewers will expect
Datasets and baselines named in the 34 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 · 9 of 34 papers
- Snow100K · 8 of 34 papers
- BSD400 · 7 of 34 papers
- WED · 7 of 34 papers
- Rain100L · 6 of 34 papers
- RESIDE · 5 of 34 papers
- DIV2K · 4 of 34 papers
- SOTS · 4 of 34 papers
Baselines
- Restormer · 15 of 34 papers
- PromptIR · 14 of 34 papers
- AirNet · 11 of 34 papers
- MPRNet · 10 of 34 papers
- NAFNet · 8 of 34 papers
- IDR · 7 of 34 papers
- TransWeather · 6 of 34 papers
- MambaIR · 6 of 34 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 95 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 7 papers
- The Hong Kong Polytechnic University · 5 papers
- Harbin Institute of Technology · 4 papers
- South China University of Technology · 4 papers
- Tsinghua University · 4 papers
- Shenzhen University · 4 papers
- Dalian University of Technology · 3 papers
- Shanghai AI Laboratory · 3 papers
- The Chinese University of Hong Kong · 3 papers
- Sun Yat-sen University · 3 papers
Use it
Download BibTeX of 33 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
- Priors for restoration networks (66 papers)
- Diffusion-based universal restoration (61 papers)
- Real-world restoration with unaligned data (40 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).