Research map / Image generation / Autoregressive and latent generation
Learning-based image quality assessment
Predict perceptual image quality without reference using learned or synthetic supervision.
62 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Autoregressive and latent generation, keyword cluster “captioning · assessment · synthetic”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
21 papers in 2023–24 and 15 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 4.4% to 1.7%.
No longer reported by papers here: TID2013 (SRCC, last in 2024); CSIQ (SRCC, last in 2024); TID2013 (PLCC, last in 2024); KonIQ-10K (SRCC, last in 2024); LIVE (SRCC, 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 · Group MAD Competition - A New Methodology to Compare Objective Image Quality Models (CVPR 2016 · 98 citations)
- 2017 · Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework (CVPR 2017 · 278 citations)
- 2018 · Generalisation in humans and deep neural networks (NeurIPS 2018 · 426 citations)
- 2019 · No-Reference Image Quality Assessment with Reinforcement Recursive List-Wise Ranking (AAAI 2019 · 17 citations)
- 2020 · MetaIQA: Deep Meta-Learning for No-Reference Image Quality Assessment (CVPR 2020 · 432 citations)
- 2021 · On Feature Normalization and Data Augmentation (CVPR 2021 · 142 citations)
- 2022 · Personalized Image Aesthetics Assessment With Rich Attributes (CVPR 2022 · 90 citations)
- 2023 · Re-IQA: Unsupervised Learning for Image Quality Assessment in the Wild (CVPR 2023 · 139 citations)
- 2024 · Boosting Image Quality Assessment through Efficient Transformer Adaptation with Local Feature Enhancement (CVPR 2024 · 48 citations)
- 2025 · Teaching Large Language Models to Regress Accurate Image Quality Scores Using Score Distribution (CVPR 2025 · 44 citations)
- 2026 · Mitigating Perception Bias: A Training-Free Approach to Enhance LMM for Image Quality Assessment (AAAI 2026 · 2 citations)
Earliest papers here
- Group MAD Competition - A New Methodology to Compare Objective Image Quality Models (CVPR 2016 · 98 citations)
- Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework (CVPR 2017 · 278 citations)
- Eigen-Distortions of Hierarchical Representations (NeurIPS 2017 · 19 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.
TID2013 · SRCC (higher is better)
10 papers, 2017–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.939 | Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework (CVPR 2017 · 278 citations) |
| 2018 | 0.945 | PieAPP: Perceptual Image-Error Assessment Through Pairwise Preference (CVPR 2018 · 332 citations) |
CSIQ · PLCC (higher is better)
9 papers, 2017–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.965 | Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework (CVPR 2017 · 278 citations) |
| 2018 | 0.975 | PieAPP: Perceptual Image-Error Assessment Through Pairwise Preference (CVPR 2018 · 332 citations) |
CSIQ · SRCC (higher is better)
8 papers, 2017–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.961 | Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework (CVPR 2017 · 278 citations) |
| 2018 | 0.973 | PieAPP: Perceptual Image-Error Assessment Through Pairwise Preference (CVPR 2018 · 332 citations) |
TID2013 · PLCC (higher is better)
8 papers, 2017–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.947 | Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework (CVPR 2017 · 278 citations) |
| 2024 | 0.948 | Transformer-Based No-Reference Image Quality Assessment via Supervised Contrastive Learning (AAAI 2024 · 45 citations) |
KonIQ-10K · SRCC (higher is better)
6 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 0.789 | Blind Predicting Similar Quality Map for Image Quality Assessment (CVPR 2018 · 106 citations) |
| 2024 | 0.934 | Blind Image Quality Assessment Based on Geometric Order Learning (CVPR 2024 · 26 citations) |
LIVE · SRCC (higher is better)
6 papers, 2022–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 0.913 | Content-Variant Reference Image Quality Assessment via Knowledge Distillation (AAAI 2022 · 47 citations) |
| 2023 | 0.98 | Data-Efficient Image Quality Assessment with Attention-Panel Decoder (AAAI 2023 · 87 citations) |
| 2024 | 0.983 | Integrating Global Context Contrast and Local Sensitivity for Blind Image Quality Assessment (ICML 2024) |
What has been tried
52 papers here report 541 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 6%, hurt 17%, 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 a local distortion extractor and injector to ViT | KADID-10k · SRCC | 0.889 -> 0.931 | Helped | Boosting Image Quality Assessment through Efficient Transformer Adaptation with Local Feature Enhancement (CVPR 2024 · 48 citations) |
| add supervised contrastive learning (SCL) | LIVE Challenge · PLCC | 0.827 -> 0.871 | Helped | Transformer-Based No-Reference Image Quality Assessment via Supervised Contrastive Learning (AAAI 2024 · 45 citations) |
| replace one-hot score labels with distribution-based soft labels | KonIQ · PLCC | 0.945 -> 0.954 | Helped | Teaching Large Language Models to Regress Accurate Image Quality Scores Using Score Distribution (CVPR 2025 · 44 citations) |
| Add quality-factor learning through the merged-factor distribution | Eight benchmarks · pAUC@FMR=1E−3 (Avg.) | 0.705 -> 0.763 | Hurt | CLIB-FIQA: Face Image Quality Assessment with Confidence Calibration (CVPR 2024 · 34 citations) |
| Average predictions across trials and subjects rather than use no average | THINGS-MEG · CLIP | 0.700 -> 0.811 | Helped | Brain decoding: toward real-time reconstruction of visual perception (ICLR 2024 · 27 citations) |
| Add FSU and PSU modules | KonIQ10K · SRCC | 0.858 -> 0.934 | Helped | Blind Image Quality Assessment Based on Geometric Order Learning (CVPR 2024 · 26 citations) |
| add Omni MoE with subject-specific parameters and Ecphory to the ViT baseline | NSD test · CLIP | 81.6% -> 96.8% | Helped | Psychometry: An Omnifit Model for Image Reconstruction from Human Brain Activity (CVPR 2024 · 20 citations) |
| replace CLIP ViT-B/32 with VGG16 | Standard IQA average · accuracy (%) | 83.6 -> 77.6 | Hurt | Toward Generalized Image Quality Assessment: Relaxing the Perfect Reference Quality Assumption (CVPR 2025 · 15 citations) |
| Use the probability matrix instead of the count matrix | LIVE · SRCC | 0.888 -> 0.974 | Helped | Adaptive Image Quality Assessment via Teaching Large Multimodal Model to Compare (NeurIPS 2024 · 10 citations) |
| Reduce generated responses from K=6 to K=4 | 8-dataset average · PLCC | 0.814 -> 0.811 | No clear effect | VisualQuality-R1: Reasoning-Induced Image Quality Assessment via Reinforcement Learning to Rank (NeurIPS 2025 · 8 citations) |
| Use probability augmentation rather than complete augmentation | SIDD · PSNR (dB) | +0.0214 dB | Helped | CutFreq: Cut-and-Swap Frequency Components for Low-Level Vision Augmentation (AAAI 2024 · 6 citations) |
| use DDIM inversion instead of the VAE alone | BRI3L · PAS | 0.00 -> 84.17 | Helped | The Art of Deception: Color Visual Illusions and Diffusion Models (CVPR 2025 · 3 citations) |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- TID2013 · 5 of 8 papers
- KADID-10K · 5 of 8 papers
- PIPAL · 4 of 8 papers
- CSIQ · 4 of 8 papers
- SPAQ · 2 of 8 papers
- LIVE · 2 of 8 papers
Baselines
- Q-Align · 3 of 8 papers
- MUSIQ · 3 of 8 papers
- VAE · 2 of 8 papers
- NIQE · 2 of 8 papers
- BRISQUE · 2 of 8 papers
- HyperIQA · 2 of 8 papers
- DBCNN · 2 of 8 papers
- CLIP-IQA+ · 2 of 8 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).
- Tsinghua University · 4 papers
- Xiamen University · 3 papers
- Ocean University of China · 3 papers
- Chinese Academy of Sciences · 3 papers
- City University of Hong Kong · 3 papers
- Hong Kong University of Science and Technology · 2 papers
- Nanyang Technological University · 2 papers
- Sony · 2 papers
- Tencent · 2 papers
- XMU · 2 papers
Use it
Download BibTeX of 37 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
- Classical generative image synthesis (147 papers)
- Retrieval and matching with generative features (140 papers)
- Synthetic training data from generators (111 papers)
Elsewhere in the topic
- High-resolution diffusion synthesis (140 papers)
- Generative latent space tooling (129 papers)
- Generalizable fake image detection (119 papers)
- Latent diffusion with learned tokenizers (114 papers)
- Compositional scene generation (105 papers)
- Next-scale and token autoregressive images (90 papers)
- Robust diffusion watermarking (79 papers)
- Neural text-to-speech synthesis (72 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).