Research map / Image generation / Super-resolution
Perceptual and real-world SR
Improve realism and semantic accuracy of super-resolution with perceptual losses and semantic guidance.
85 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Super-resolution, keyword cluster “image super · attention · resolution diffusion”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 6 tracked benchmarks.
23 papers in 2023–24 and 39 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 19.7% to 23.6%.
Open questions the data points to
- Manga109 ×4 (PSNR): best went from 31.33 (up to 2024) to 31.95 in 2025 — TSP-Mamba: The Travelling Salesman Problem Meets Mamba for Image Super-resolution and Beyond (CVPR 2025 · 5 citations)
What looks settled
- Urban100 ×4 (PSNR): best up to 2024 27.54, best in 2025–26 27.29 from 2 papers, at most 0.2% better, if at all.
- Set14 ×4 (PSNR): best up to 2024 28.94, best in 2025–26 29.05 from 2 papers, less than 1% better.
- Set5 ×4 (PSNR): best up to 2024 32.67, best in 2025–26 32.71 from 2 papers, at most 0.2% better, if at all.
No longer reported by papers here: B100 ×4 (PSNR, last in 2024); Set14 ×2 (PSNR, last in 2023).
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 · Seven Ways to Improve Example-Based Single Image Super Resolution (CVPR 2016 · 487 citations)
- 2017 · Simultaneous Super-Resolution and Cross-Modality Synthesis of 3D Medical Images Using Weakly-Supervised Joint Convolutional Sparse Coding (CVPR 2017 · 216 citations)
- 2018 · Super-FAN: Integrated Facial Landmark Localization and Super-Resolution of Real-World Low Resolution Faces in Arbitrary Poses With GANs (CVPR 2018 · 379 citations)
- 2019 · Zoom to Learn, Learn to Zoom (CVPR 2019 · 277 citations)
- 2020 · Learning Light Field Angular Super-Resolution via a Geometry-Aware Network (AAAI 2020 · 141 citations)
- 2021 · End-to-End Learning for Joint Image Demosaicing, Denoising and Super-Resolution (CVPR 2021 · 88 citations)
- 2022 · Local Texture Estimator for Implicit Representation Function (CVPR 2022 · 235 citations)
- 2023 · Cascaded Local Implicit Transformer for Arbitrary-Scale Super-Resolution (CVPR 2023 · 73 citations)
- 2024 · CoSeR: Bridging Image and Language for Cognitive Super-Resolution (CVPR 2024 · 51 citations)
- 2025 · The Power of Context: How Multimodality Improves Image Super-Resolution (CVPR 2025 · 8 citations)
- 2026 · Coloring the Noise: Adversarial Sobolev Alignment for Faithful Image Super Resolution (ICML 2026 · 0 citations)
Earliest papers here
- Seven Ways to Improve Example-Based Single Image Super Resolution (CVPR 2016 · 487 citations)
- Towards Domain Adaptive Vehicle Detection in Satellite Image by Supervised Super-Resolution Transfer (AAAI 2016 · 19 citations)
- Simultaneous Super-Resolution and Cross-Modality Synthesis of 3D Medical Images Using Weakly-Supervised Joint Convolutional Sparse Coding (CVPR 2017 · 216 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 ×4 · PSNR (higher is better)
8 papers, 2019–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 25.7 | Multigrid Backprojection Super–Resolution and Deep Filter Visualization (AAAI 2019 · 26 citations) |
| 2022 | 27.12 | Efficient Non-local Contrastive Attention for Image Super-resolution (AAAI 2022 · 105 citations) |
| 2023 | 27.54 | Deep Arbitrary-Scale Image Super-Resolution via Scale-Equivariance Pursuit (CVPR 2023 · 29 citations) |
Set14 ×4 · PSNR (higher is better)
7 papers, 2019–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 28.43 | Multigrid Backprojection Super–Resolution and Deep Filter Visualization (AAAI 2019 · 26 citations) |
| 2022 | 28.94 | Efficient Non-local Contrastive Attention for Image Super-resolution (AAAI 2022 · 105 citations) |
| 2025 | 29.05 | TSP-Mamba: The Travelling Salesman Problem Meets Mamba for Image Super-resolution and Beyond (CVPR 2025 · 5 citations) |
Manga109 ×4 · PSNR (higher is better)
5 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 30.07 | Multigrid Backprojection Super–Resolution and Deep Filter Visualization (AAAI 2019 · 26 citations) |
| 2022 | 31.33 | Efficient Non-local Contrastive Attention for Image Super-resolution (AAAI 2022 · 105 citations) |
| 2025 | 31.95 | TSP-Mamba: The Travelling Salesman Problem Meets Mamba for Image Super-resolution and Beyond (CVPR 2025 · 5 citations) |
Set5 ×4 · PSNR (higher is better)
5 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 32.67 | Efficient Non-local Contrastive Attention for Image Super-resolution (AAAI 2022 · 105 citations) |
| 2025 | 32.71 | TSP-Mamba: The Travelling Salesman Problem Meets Mamba for Image Super-resolution and Beyond (CVPR 2025 · 5 citations) |
B100 ×4 · PSNR (higher is better)
4 papers, 2022–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 27.82 | Efficient Non-local Contrastive Attention for Image Super-resolution (AAAI 2022 · 105 citations) |
Set14 ×2 · PSNR (higher is better)
4 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 33.27 | Multigrid Backprojection Super–Resolution and Deep Filter Visualization (AAAI 2019 · 26 citations) |
| 2022 | 34.17 | Efficient Non-local Contrastive Attention for Image Super-resolution (AAAI 2022 · 105 citations) |
What has been tried
61 papers here report 648 measured changes (a component added, removed or swapped, compared with a baseline): helped 75%, no clear effect 5%, hurt 17%, 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 multi-token cognitive encoder rather than omit cognitive encoding | ImageNet Test2000 · FID | 23.18 -> 20.27 | Helped | CoSeR: Bridging Image and Language for Cognitive Super-Resolution (CVPR 2024 · 51 citations) |
| add cycle-consistent loss | Chest dataset x5 · SNLI | 1.1714 -> 0.6134 | Helped | CycleINR: Cycle Implicit Neural Representation for Arbitrary-Scale Volumetric Super-Resolution of Medical Data (CVPR 2024 · 21 citations) |
| add diffusion-based detail enhancement after IC-FSRNet | CelebAMaskHQ test · NIQE | 10.585 -> 6.129 | Helped | Low-Light Face Super-resolution via Illumination, Structure, and Texture Associated Representation (AAAI 2024 · 16 citations) |
| use EDSR-baseline rather than UNet as the latent generator | DIV2K validation 4× · LPIPS | 0.119 -> 0.121 | Hurt | Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · 14 citations) |
| Enable classifier-free guidance with step size 0.5 instead of disabling it with step size 1.0 | NoCaps · LR PSNR (dB) | 49.67 -> 24.31 | Hurt | Text-guided Explorable Image Super-resolution (CVPR 2024 · 14 citations) |
| add meta-learning instead of directly embedding coordinates | COZ test ×2, ×3, ×4, ×5, ×6 · PSNR (dB) | 28.77 / 26.58 / 25.15 / 24.34 / 23.23 -> 28.86 / 26.66 / 25.22 / 24.39 / 23.29 | Helped | Continuous Optical Zooming: A Benchmark for Arbitrary-Scale Image Super-Resolution in Real World (CVPR 2024 · 13 citations) |
| add the Multimodal Latent Connector | DIV2K-Val-100 1024 × 1024 · LPIPS | 0.3929 -> 0.2810 | Helped | The Power of Context: How Multimodality Improves Image Super-Resolution (CVPR 2025 · 8 citations) |
| Update all scales jointly rather than iteratively updating each level separately | Text-conditioned zoom sequences · cross-scale consistency | Helped | Generative Powers of Ten (CVPR 2024 · 7 citations) | |
| Add a decoder without an encoder | DIV2K-val · DISTS | 0.0526 -> 0.0521 | Helped | Auto-Encoded Supervision for Perceptual Image Super-Resolution (CVPR 2025 · 6 citations) |
| use 1,000 instead of 3 sampling steps | Cityscapes · FID | 56.07 -> 42.70 | Helped | Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators (CVPR 2025 · 6 citations) |
| Use generative priors with MPF, POD, and 3DA instead of the residual-block control | CREMI C 16× · LPIPS | 0.5168 -> 0.4790 | Helped | Learning Large-Factor EM Image Super-Resolution with Generative Priors (CVPR 2024 · 6 citations) |
| use negative cosine distance rather than L1 or L2 for scan-path cost | Urban100 ×4 · PSNR | 26.55 -> 26.61; 26.54 -> 26.61 | Helped | TSP-Mamba: The Travelling Salesman Problem Meets Mamba for Image Super-resolution and Beyond (CVPR 2025 · 5 citations) |
What reviewers will expect
Datasets and baselines named in the 25 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 · 11 of 25 papers
- RealSR · 8 of 25 papers
- DRealSR · 7 of 25 papers
- DF2K · 4 of 25 papers
- LSDIR · 4 of 25 papers
- ImageNet · 4 of 25 papers
- Set5 · 3 of 25 papers
- Set14 · 3 of 25 papers
Baselines
- SeeSR · 8 of 25 papers
- Real-ESRGAN · 5 of 25 papers
- SwinIR · 5 of 25 papers
- StableSR · 4 of 25 papers
- ResShift · 4 of 25 papers
- PASD · 4 of 25 papers
- DiffBIR · 4 of 25 papers
- SUPIR · 4 of 25 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 62 papers; from OpenAlex and the papers' affiliations).
- Harbin Institute of Technology · 4 papers
- Google · 4 papers
- The Hong Kong Polytechnic University · 4 papers
- Shanghai Jiao Tong University · 3 papers
- Huawei · 3 papers
- Tsinghua University · 3 papers
- Chinese Academy of Sciences · 3 papers
- MediaTek Inc · 2 papers
- National Tsing Hua University · 2 papers
- University of Science and Technology of China · 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
- Efficient lightweight super-resolution (101 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).