Research map / Image generation / Super-resolution
Efficient lightweight super-resolution
Reduce super-resolution cost with lightweight networks, selective processing, and mixtures of experts.
101 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Super-resolution, keyword cluster “image super · attention · resolution diffusion”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Leveling off Several papers from 2025–26 still report on 4 tracked benchmarks, but the best result has barely moved since 2024.
27 papers in 2023–24 and 43 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 23.1% to 26.1%.
What looks settled
- Manga109 ×4 (PSNR): best up to 2024 32.5, best in 2025–26 32.63 from 2 papers, less than 1% better.
- Set14 ×4 (PSNR): best up to 2024 29.23, best in 2025–26 29.29 from 2 papers, less than 1% better.
- Set5 ×4 (PSNR): best up to 2024 33.12, best in 2025–26 33.15 from 2 papers, at most 0.2% better, if at all.
- Urban100 ×4 (PSNR): best up to 2024 28.6, best in 2025–26 28.2 from 2 papers, at most 0.2% better, if at all.
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
- 2017 · Hyperspectral Image Super-Resolution via Non-Local Sparse Tensor Factorization (CVPR 2017 · 341 citations)
- 2018 · Unsupervised Sparse Dirichlet-Net for Hyperspectral Image Super-Resolution (CVPR 2018 · 267 citations)
- 2019 · Learning Parallax Attention for Stereo Image Super-Resolution (CVPR 2019 · 331 citations)
- 2020 · Unsupervised Adaptation Learning for Hyperspectral Imagery Super-Resolution (CVPR 2020 · 137 citations)
- 2021 · Scene Text Telescope: Text-Focused Scene Image Super-Resolution (CVPR 2021 · 159 citations)
- 2022 · Discrete Cosine Transform Network for Guided Depth Map Super-Resolution (CVPR 2022 · 142 citations)
- 2023 · Activating More Pixels in Image Super-Resolution Transformer (CVPR 2023 · 1,108 citations)
- 2024 · CAMixerSR: Only Details Need More "Attention" (CVPR 2024 · 76 citations)
- 2025 · CATANet: Efficient Content-Aware Token Aggregation for Lightweight Image Super-Resolution (CVPR 2025 · 62 citations)
- 2026 · MFmamba: A Multi-function Network for Panchromatic Image Resolution Restoration Based on State-Space Model (AAAI 2026 · 3 citations)
Earliest papers here
- Hyperspectral Image Super-Resolution via Non-Local Sparse Tensor Factorization (CVPR 2017 · 341 citations)
- Amortised MAP Inference for Image Super-resolution (ICLR 2017 · 322 citations)
- Unsupervised Sparse Dirichlet-Net for Hyperspectral Image Super-Resolution (CVPR 2018 · 267 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.
Manga109 ×4 · PSNR (higher is better)
9 papers, 2023–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 32.48 | Activating More Pixels in Image Super-Resolution Transformer (CVPR 2023 · 1,108 citations) |
| 2024 | 32.5 | Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · 24 citations) |
| 2025 | 32.63 | Progressive Focused Transformer for Single Image Super-Resolution (CVPR 2025 · 47 citations) |
Set14 ×4 · PSNR (higher is better)
9 papers, 2023–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 29.23 | Activating More Pixels in Image Super-Resolution Transformer (CVPR 2023 · 1,108 citations) |
| 2025 | 29.29 | Progressive Focused Transformer for Single Image Super-Resolution (CVPR 2025 · 47 citations) |
Set5 ×4 · PSNR (higher is better)
9 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 33.07 | Activating More Pixels in Image Super-Resolution Transformer (CVPR 2023 · 1,108 citations) |
| 2024 | 33.12 | Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · 24 citations) |
| 2025 | 33.15 | Progressive Focused Transformer for Single Image Super-Resolution (CVPR 2025 · 47 citations) |
Urban100 ×4 · PSNR (higher is better)
9 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 28.6 | Activating More Pixels in Image Super-Resolution Transformer (CVPR 2023 · 1,108 citations) |
B100 ×4 · PSNR (higher is better)
5 papers, 2023–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 27.79 | From Coarse to Fine: Hierarchical Pixel Integration for Lightweight Image Super-resolution (AAAI 2023 · 26 citations) |
| 2024 | 28 | Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · 24 citations) |
BSD100 ×4 · PSNR (higher is better)
5 papers, 2023–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 28 | Activating More Pixels in Image Super-Resolution Transformer (CVPR 2023 · 1,108 citations) |
| 2024 | 28.17 | CFAT: Unleashing Triangular Windows for Image Super-resolution (CVPR 2024 · 69 citations) |
What has been tried
72 papers here report 754 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 5%, hurt 16%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add deformable offsets | Urban100 ×4 · PSNR | 26.60 -> 26.63 | Helped | CAMixerSR: Only Details Need More "Attention" (CVPR 2024 · 76 citations) |
| add CWAB | BSD100 ×4 · PSNR | 27.65dB -> 27.75dB | Helped | CFAT: Unleashing Triangular Windows for Image Super-resolution (CVPR 2024 · 69 citations) |
| add IASA to the LRSA-only baseline | Urban100 ×4 · PSNR | 26.46 -> 26.85 | Helped | CATANet: Efficient Content-Aware Token Aggregation for Lightweight Image Super-Resolution (CVPR 2025 · 62 citations) |
| Add Progressive Attention (Hadamard inheritance) | Set5 ×4 · PSNR | 32.28 -> 32.35 | Helped | Progressive Focused Transformer for Single Image Super-Resolution (CVPR 2025 · 47 citations) |
| replace LPIPS with DISTS perceptual loss | BSD100 · PSNR | 24.506 -> 24.812 | Helped | Training Generative Image Super-Resolution Models by Wavelet-Domain Losses Enables Better Control of Artifacts (CVPR 2024 · 45 citations) |
| add the symmetric auto-encoder decoder before the MLP | Decoder reconstruction 16 →128 · PSNR | 22.96 -> 35.28 | Helped | Arbitrary-Scale Image Generation and Upsampling using Latent Diffusion Model and Implicit Neural Decoder (CVPR 2024 · 33 citations) |
| Add Gaussian splatting for 8 channels, with bicubic upsampling on the remaining 56, instead of bicubic upsampling on all 64 channels | DIV2K100 ×2, ×3, and ×4 · PSNR (dB) | 34.4128 -> 34.7134; 30.8311 -> 31.0010; 28.8788 -> 29.0281 | Helped | GaussianSR: High Fidelity 2D Gaussian Splatting for Arbitrary-Scale Image Super-Resolution (AAAI 2025 · 31 citations) |
| Add RG-SA to the L-SA baseline | Urban100 ×2 · PSNR | 33.43 -> 33.52 | Helped | Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · 24 citations) |
| Use bilateral spatial and spectral super-resolution branches instead of only the spectral branch | CAVE · PSNR | 41.0376 -> 43.9264 | Helped | S2CycleDiff: Spatial-Spectral-Bilateral Cycle-Diffusion Framework for Hyperspectral Image Super-resolution (AAAI 2024 · 21 citations) |
| Add hierarchical encoding | Urban100 ×4 · PSNR (dB) | 26.38 -> 26.51 | Helped | HIIF: Hierarchical Encoding based Implicit Image Function for Continuous Super-resolution (CVPR 2025 · 18 citations) |
| add MMRD | Pavia Center · PSNR | 43.6667 -> 45.3475 | Helped | Learning Multi-Modal Cross-Scale Deformable Transformer Network for Unregistered Hyperspectral Image Super-resolution (AAAI 2024 · 16 citations) |
| Add ImageNet pre-training | BDD100K · PSNR | 29.56 to 29.91 | Helped | Transformer-Based Selective Super-resolution for Efficient Image Refinement (AAAI 2024 · 15 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 |
|---|---|---|---|---|
| ResShift | DRealSR | PSNR | 26.19 – 28.46 | 2 papers, e.g. ICLR 2025 and NeurIPS 2025 |
What reviewers will expect
Datasets and baselines named in the 18 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 · 5 of 18 papers
- RealSR · 3 of 18 papers
- DRealSR · 2 of 18 papers
Baselines
- Bicubic · 4 of 18 papers
- LIIF · 3 of 18 papers
- SwinIR · 3 of 18 papers
- HAT · 3 of 18 papers
- MetaSR · 2 of 18 papers
- IMDN · 2 of 18 papers
- EDSR · 2 of 18 papers
- RCAN · 2 of 18 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 70 papers; from OpenAlex and the papers' affiliations).
- The Chinese University of Hong Kong · 4 papers
- Peking University · 3 papers
- Swiss Federal Institute of Technology · 3 papers
- Nanjing University · 2 papers
- ByteDance · 2 papers
- Xidian University · 2 papers
- Beijing University of Posts and Telecommunications · 2 papers
- Ocean University of China · 2 papers
- Harbin Institute of Technology · 2 papers
- The Hong Kong Polytechnic University · 2 papers
Use it
Download BibTeX of 38 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
- Perceptual and real-world SR (85 papers)
- CNN and transformer SR architectures (62 papers)
- Diffusion-based super-resolution (85 papers)
Elsewhere in the topic
- Real-world super-resolution (72 papers)
- Blind super-resolution (24 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).