Research map / Image generation / Autoregressive and latent generation
Retrieval and matching with generative features
Apply generative or diffusion features to re-ranking, matching, and metric learning.
140 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 Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
42 papers in 2023–24 and 56 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 8.9% to 6.2%.
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 · Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · 944 citations)
- 2017 · Comparative Evaluation of Hand-Crafted and Learned Local Features (CVPR 2017 · 321 citations)
- 2018 · Leveraging Unlabeled Data for Crowd Counting by Learning to Rank (CVPR 2018 · 323 citations)
- 2019 · Lending Orientation to Neural Networks for Cross-View Geo-Localization (CVPR 2019 · 332 citations)
- 2020 · Tangent Images for Mitigating Spherical Distortion (CVPR 2020 · 141 citations)
- 2021 · VIGOR: Cross-View Image Geo-Localization Beyond One-to-One Retrieval (CVPR 2021 · 226 citations)
- 2022 · Effective Conditioned and Composed Image Retrieval Combining CLIP-Based Features (CVPR 2022 · 161 citations)
- 2023 · DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data (NeurIPS 2023 · 46 citations)
- 2024 · OmniGlue: Generalizable Feature Matching with Foundation Model Guidance (CVPR 2024 · 86 citations)
- 2025 · U-KAN Makes Strong Backbone for Medical Image Segmentation and Generation (AAAI 2025 · 308 citations)
- 2026 · Virtual Multiplex Staining for Histological Images Using a Marker-Wise Conditioned Diffusion Model (AAAI 2026 · 2 citations)
Earliest papers here
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · 944 citations)
- Progressive Feature Matching With Alternate Descriptor Selection and Correspondence Enrichment (CVPR 2016 · 21 citations)
- Using Spatial Order to Boost the Elimination of Incorrect Feature Matches (CVPR 2016 · 3 citations)
What has been tried
111 papers here report 1,114 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 5%, hurt 21%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use 3 KAN layers instead of 1 | Segmentation ablation · IoU | 64.20 -> 65.26 | Helped | U-KAN Makes Strong Backbone for Medical Image Segmentation and Generation (AAAI 2025 · 308 citations) |
| remove position guidance | GSO-Hard · pose accuracy @5° | 11.0 -> 10.0 | Hurt | OmniGlue: Generalizable Feature Matching with Foundation Model Guidance (CVPR 2024 · 86 citations) |
| train LoFTR from scratch on synthetic RGB-IR | Synthetic RGB-IR · AUC@10° | 12.58 -> 23.63 | Helped | MINIMA: Modality Invariant Image Matching (CVPR 2025 · 44 citations) |
| add freezing of the last CLIP layer after pretraining | GeoGuessr holdout, 5,000 Street View locations · median error (km) | 45.47 -> 44.35 | Helped | PIGEON: Predicting Image Geolocations (CVPR 2024 · 26 citations) |
| Add semantic-shift decomposition to the baseline | CIRR validation · Recall@1 | 42.62 -> 45.13 | Helped | Decomposing Semantic Shifts for Composed Image Retrieval (AAAI 2024 · 23 citations) |
| Use APSC-Net with MIML training data rather than without MIML | CASIAv1 · IoU | .799 -> .810 | Helped | Towards Modern Image Manipulation Localization: A Large-Scale Dataset and Novel Methods (CVPR 2024 · 18 citations) |
| Add ImageNet pretraining | TexTile test set · Error | 0.459 -> 0.064 | Helped | TexTile: A Differentiable Metric for Texture Tileability (CVPR 2024 · 17 citations) |
| replace abundance-space generation with original HSI-cube-space generation | Chikusei HSI · generation quality | Hurt | Unmixing Before Fusion: A Generalized Paradigm for Multi-Source-based Hyperspectral Image Synthesis (CVPR 2024 · 15 citations) | |
| Vary the number of nearest neighbors K | Six VPR datasets · AUC-PR | Helped | On the Estimation of Image-matching Uncertainty in Visual Place Recognition (CVPR 2024 · 13 citations) | |
| add LIPN instance pre-screening to DMIN | Camelyon16 test · inference time | 23.46 -> 16.75 | Helped | Fast and Accurate Gigapixel Pathological Image Classification with Hierarchical Distillation Multi-Instance Learning (CVPR 2025 · 11 citations) |
| Add Adaptive Sampling to the Prompt Predictor | SA-1B · mIoU | 57.2 -> 72.8 | Helped | AoP-SAM: Automation of Prompts for Efficient Segmentation (AAAI 2025 · 8 citations) |
| Add DPE instead of ordinary patch embedding | HBC, HER2+, and cSCC · PCC (%) | 47.13 -> 48.07; 43.10 -> 44.17; 49.35 -> 50.50 | Helped | M2OST: Many-to-one Regression for Predicting Spatial Transcriptomics from Digital Pathology Images (AAAI 2025 · 8 citations) |
What reviewers will expect
Datasets and baselines named in the 38 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- BUSI · 2 of 38 papers
- GlaS · 2 of 38 papers
- HCP · 2 of 38 papers
- FashionIQ · 2 of 38 papers
- CIRR · 2 of 38 papers
- CUB-200 · 2 of 38 papers
- HEST-1k · 2 of 38 papers
- HER2ST · 2 of 38 papers
Baselines
- UNet · 3 of 38 papers
- ST-Net · 3 of 38 papers
- HisToGene · 3 of 38 papers
- BLEEP · 3 of 38 papers
- Att-UNet · 3 of 38 papers
- Mathpix · 2 of 38 papers
- ResNet50 · 2 of 38 papers
- ViT-B/16 · 2 of 38 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 98 papers; from OpenAlex and the papers' affiliations).
- Massachusetts Institute of Technology · 4 papers
- Zhejiang University · 3 papers
- South China University of Technology · 3 papers
- City University of Hong Kong · 3 papers
- The Chinese University of Hong Kong · 3 papers
- Adobe · 3 papers
- Google · 3 papers
- The University of Sydney · 2 papers
- Wuhan University · 2 papers
- Oregon State 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
- Classical generative image synthesis (147 papers)
- Synthetic training data from generators (111 papers)
- Learning-based image quality assessment (62 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).