Research map / Image generation / Super-resolution
Real-world super-resolution
Handle ill-posed super-resolution with degradation modeling, adaptation, and frequency-aware representations.
72 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Super-resolution, keyword cluster “lr · hr · degradation”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
22 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 18.8% to 6.7%.
No longer reported by papers here: Urban100 (PSNR, last in 2023); CUFED5 (PSNR, last in 2022); Manga109 (PSNR, last in 2023); Sun80 (PSNR, last in 2022).
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 · Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (CVPR 2016 · 7,310 citations)
- 2017 · Attention-Aware Face Hallucination via Deep Reinforcement Learning (CVPR 2017 · 240 citations)
- 2018 · Learning a Single Convolutional Super-Resolution Network for Multiple Degradations (CVPR 2018 · 1,151 citations)
- 2019 · Image Super-Resolution by Neural Texture Transfer (CVPR 2019 · 385 citations)
- 2020 · Learning Texture Transformer Network for Image Super-Resolution (CVPR 2020 · 970 citations)
- 2021 · MASA-SR: Matching Acceleration and Spatial Adaptation for Reference-Based Image Super-Resolution (CVPR 2021 · 176 citations)
- 2022 · Best-Buddy GANs for Highly Detailed Image Super-resolution (AAAI 2022 · 84 citations)
- 2023 · Super-Resolution Neural Operator (CVPR 2023 · 77 citations)
- 2024 · SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024 · 189 citations)
- 2025 · Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution (AAAI 2025 · 2 citations)
- 2026 · The False Promise of Zero-Shot Super-Resolution in Machine-Learned Operators (ICLR 2026 · 0 citations)
Earliest papers here
- Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (CVPR 2016 · 7,310 citations)
- Learning Parametric Sparse Models for Image Super-Resolution (NeurIPS 2016 · 7 citations)
- Attention-Aware Face Hallucination via Deep Reinforcement Learning (CVPR 2017 · 240 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.
Urban100 · PSNR (higher is better)
8 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 25.09 | Image Super-Resolution by Neural Texture Transfer (CVPR 2019 · 385 citations) |
| 2020 | 26.71 | Robust Reference-Based Super-Resolution With Similarity-Aware Deformable Convolution (CVPR 2020 · 126 citations) |
CUFED5 · PSNR (higher is better)
6 papers, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 25.61 | Image Super-Resolution by Neural Texture Transfer (CVPR 2019 · 385 citations) |
| 2020 | 26.78 | Robust Reference-Based Super-Resolution With Similarity-Aware Deformable Convolution (CVPR 2020 · 126 citations) |
| 2021 | 28.24 | Robust Reference-Based Super-Resolution via C2-Matching (CVPR 2021 · 5 citations) |
| 2022 | 28.64 | Task Decoupled Framework for Reference-Based Super-Resolution (CVPR 2022 · 37 citations) |
Manga109 · PSNR (higher is better)
5 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 30.09 | Learning Texture Transformer Network for Image Super-Resolution (CVPR 2020 · 970 citations) |
| 2021 | 30.47 | Robust Reference-Based Super-Resolution via C2-Matching (CVPR 2021 · 5 citations) |
| 2022 | 31.23 | Task Decoupled Framework for Reference-Based Super-Resolution (CVPR 2022 · 37 citations) |
Sun80 · PSNR (higher is better)
5 papers, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 27.59 | Image Super-Resolution by Neural Texture Transfer (CVPR 2019 · 385 citations) |
| 2020 | 30.02 | Learning Texture Transformer Network for Image Super-Resolution (CVPR 2020 · 970 citations) |
| 2021 | 30.18 | Robust Reference-Based Super-Resolution via C2-Matching (CVPR 2021 · 5 citations) |
| 2022 | 30.31 | Task Decoupled Framework for Reference-Based Super-Resolution (CVPR 2022 · 37 citations) |
What has been tried
65 papers here report 672 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 4%, hurt 16%, 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 LR latent to the initial sampling noise | smooth regions such as the sky · visual artifacts | Helped | SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024 · 189 citations) | |
| Add GCM to the baseline | Middlebury ×16 · RMSE (cm) | 0.11cm lower | Helped | SGNet: Structure Guided Network via Gradient-Frequency Awareness for Depth Map Super-resolution (AAAI 2024 · 49 citations) |
| Add 40 same-sensor images to fine-tuning (50 images total) | RealSR Nikon · LPIPS | 0.2536 -> 0.2571 | Mixed | Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · 30 citations) |
| Use RSD instead of SDS for upscaling | NeRF-Synthetic and LLFF · NIQE | 6.273 -> 5.386; 5.903 -> 5.544 | Helped | DiSR-NeRF: Diffusion-Guided View-Consistent Super-Resolution NeRF (CVPR 2024 · 23 citations) |
| replace traditional dense feature matching with center and corner warping | DuSR-Real · PSNR | 27.03 -> 27.66 | Helped | KeDuSR: Real-World Dual-Lens Super-Resolution via Kernel-Free Matching (AAAI 2024 · 12 citations) |
| remove degradation dictionary Ddeg | AIM-RWSR · PSNR | 22.46 -> 22.40 | Hurt | Learning Coupled Dictionaries from Unpaired Data for Image Super-Resolution (CVPR 2024 · 6 citations) |
| Remove RGB guidance | NYU v2 · RMSE | 1.20 -> 1.92 (×4); 2.45 -> 3.91 (×8); 4.69 -> 7.62 (×16) | Hurt | Intrinsic Phase-Preserving Networks for Depth Super Resolution (AAAI 2024 · 4 citations) |
| use UDDM fusion with w = 0.5 instead of DDM alone (w = 0.0) | RealSR and DRealSR · PSNR | RealSR: 26.10 -> 26.63; DRealSR: 27.43 -> 28.03 | Helped | Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution (AAAI 2025 · 2 citations) |
| Divide the image into blocks using one shared scaling rate | Test4K · PSNR | 33.15 -> 33.08 | Hurt | Block-Based Multi-Scale Image Rescaling (AAAI 2025 · 1 citation) |
| Use SRFlow instead of PULSE as the upscaling function | FFHQ, 4× real-world downscaling · IDA-RD | PULSE: Bicubic 0.171±0.015, Bilinear 0.164±0.015, N.N. 0.254±0.018; SRFlow: Bicubic 0.110±0.145, Bilinear 0.090±0.067, N | Mixed | Deep Generative Model based Rate-Distortion for Image Downscaling Assessment (CVPR 2024 · 1 citation) |
| Use SwinIR as the backbone instead of SRResNet | Set14-SE · AR | 0.12 -> 0.67 | Helped | SEAL: A Framework for Systematic Evaluation of Real-World Super-Resolution (ICLR 2024 · 1 citation) |
| add PBR consistency loss | PBR mesh collection, ×4 SR · Renderings PSNR | 26.572 -> 28.001 | Helped | PBR-SR: Mesh PBR Texture Super Resolution from 2D Image Priors (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 5 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 · 2 of 5 papers
- Flickr2K · 2 of 5 papers
Baselines
- SwinIR · 2 of 5 papers
- StableSR · 2 of 5 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 33 papers; from OpenAlex and the papers' affiliations).
- Nanjing University of Science and Technology · 3 papers
- Huawei · 3 papers
- The Hong Kong Polytechnic University · 3 papers
- Shanghai Jiao Tong University · 2 papers
- Hunan University · 2 papers
- OPPO Research Institute · 2 papers
- Hong Kong University of Science and Technology · 2 papers
- University of Science and Technology of China · 2 papers
- National University of Singapore · 2 papers
- Chinese Academy of Sciences · 2 papers
Use it
Download BibTeX of 39 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
- Blind super-resolution (24 papers)
Elsewhere in the topic
- Efficient lightweight super-resolution (101 papers)
- Perceptual and real-world SR (85 papers)
- Diffusion-based super-resolution (85 papers)
- CNN and transformer SR architectures (62 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).