Research map / Vision-language and multimodal / Cross-modal learning
Cross-modal hashing
Learns compact binary codes for efficient cross-modal retrieval.
45 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Cross-modal learning, keyword cluster “cross · retrieval · image text”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
5 papers in 2023–24 and 25 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 1.7% to 3.2%.
No longer reported by papers here: MIRFlickr, image→text, 16 bits (MAP, last in 2020); NUS-WIDE, image→text, 16 bits (MAP, last in 2020); NUS-WIDE, text→image, 16 bits (MAP, last in 2020).
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 · Online Cross-Modal Hashing for Web Image Retrieval (AAAI 2016 · 109 citations)
- 2017 · Deep Cross-Modal Hashing (CVPR 2017 · 898 citations)
- 2018 · Unsupervised Generative Adversarial Cross-Modal Hashing (AAAI 2018 · 223 citations)
- 2019 · Coupled CycleGAN: Unsupervised Hashing Network for Cross-Modal Retrieval (AAAI 2019 · 122 citations)
- 2020 · Creating Something From Nothing: Unsupervised Knowledge Distillation for Cross-Modal Hashing (CVPR 2020 · 138 citations)
- 2021 · Deep Graph-neighbor Coherence Preserving Network for Unsupervised Cross-modal Hashing (AAAI 2021 · 197 citations)
- 2022 · Online Enhanced Semantic Hashing: Towards Effective and Efficient Retrieval for Streaming Multi-Modal Data (AAAI 2022 · 14 citations)
- 2023 · Long-Tail Cross Modal Hashing (AAAI 2023 · 15 citations)
- 2024 · Self-Supervised Multi-Modal Knowledge Graph Contrastive Hashing for Cross-Modal Search (AAAI 2024 · 23 citations)
- 2025 · Vision-guided Text Mining for Unsupervised Cross-modal Hashing with Community Similarity Quantization (AAAI 2025 · 10 citations)
- 2026 · Ambiguity-Tolerant Cross-Modal Hashing with Partial Labels (AAAI 2026 · 4 citations)
Earliest papers here
- Online Cross-Modal Hashing for Web Image Retrieval (AAAI 2016 · 109 citations)
- Collaborative Quantization for Cross-Modal Similarity Search (CVPR 2016 · 63 citations)
- Deep Cross-Modal Hashing (CVPR 2017 · 898 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.
MIRFlickr, image→text, 16 bits · MAP (higher is better)
4 papers, 2017–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.713 | Pairwise Relationship Guided Deep Hashing for Cross-Modal Retrieval (AAAI 2017 · 344 citations) |
| 2019 | 0.772 | Ranking-Based Deep Cross-Modal Hashing (AAAI 2019 · 70 citations) |
NUS-WIDE, image→text, 16 bits · MAP (higher is better)
4 papers, 2017–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.635 | Pairwise Relationship Guided Deep Hashing for Cross-Modal Retrieval (AAAI 2017 · 344 citations) |
NUS-WIDE, text→image, 16 bits · MAP (higher is better)
4 papers, 2017–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.681 | Pairwise Relationship Guided Deep Hashing for Cross-Modal Retrieval (AAAI 2017 · 344 citations) |
What has been tried
27 papers here report 242 measured changes (a component added, removed or swapped, compared with a baseline): helped 84%, no clear effect 2%, hurt 11%, 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 the multimodal knowledge graph and graph attention model | MSCOCO I→T, 128 bits · MAP | 0.823 -> 0.866 | Helped | Self-Supervised Multi-Modal Knowledge Graph Contrastive Hashing for Cross-Modal Search (AAAI 2024 · 23 citations) |
| Add community detection to the fused similarity matrix | MIRFLICKR-25K, I → T, 32 bits · mAP | 0.752 -> 0.754 | No clear effect | Vision-guided Text Mining for Unsupervised Cross-modal Hashing with Community Similarity Quantization (AAAI 2025 · 10 citations) |
| Add the GAT classifier | MIRFlickr25K I→T 64bit · mAP | 0.8464 -> 0.8637 | Helped | An End-To-End Graph Attention Network Hashing for Cross-Modal Retrieval (NeurIPS 2024 · 8 citations) |
| Add TAAP, AGSF, and PACL to the CLIP-feature-and-hash baseline | MIRFLICKR-25K, NUS-WIDE, and MS COCO I2T, 32bit · mAP | 0.8590 -> 0.9960; 0.7377 -> 0.9759; 0.7594 -> 0.8782 | Helped | PromptHash:Affinity-Prompted Collaborative Cross-Modal Learning for Adaptive Hashing Retrieval (CVPR 2025 · 7 citations) |
| add evidence correction loss Lkl | MIRFLICKR25K and IAPR TC-12, image-to-text and text-to-image · mAP@ALL | 0.823 -> 0.853; 0.798 -> 0.831; 0.513 -> 0.590; 0.522 -> 0.600 | Helped | Deep Evidential Hashing for Trustworthy Cross-Modal Retrieval (AAAI 2025 · 6 citations) |
| Add log-likelihood similarity preservation | MIRFlickr-25K, IAPR TC-12, and NUS-WIDE; I→T and T→I, 128 bits · mAP | 0.7981 -> 0.8041; 0.6253 -> 0.6499; 0.7991 -> 0.8015; 0.8532 -> 0.8574; 0.6598 -> 0.6810; 0.8212 -> 0.8311 | Helped | Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval (AAAI 2025 · 6 citations) |
| add the protein structure encoder to the language model | DNA 129 · F1 | 0.511 -> 0.574 | Helped | CrossBind: Collaborative Cross-Modal Identification of Protein Nucleic-Acid-Binding Residues (AAAI 2024 · 6 citations) |
| add autoencoder reconstruction loss | NUS-WIDE, MIRFlickr and MSCOCO, I→T and T→I, 128 bits · mAP@50 | 0.856 -> 0.865; 0.832 -> 0.837; 0.936 -> 0.942; 0.914 -> 0.917; 0.909 -> 0.916; 0.924 -> 0.931 | Helped | Statistical Model-driven Similarity Hashing: Bridging Modalities for Efficient Unsupervised Retrieval (AAAI 2025 · 4 citations) |
| Add RBF-kernelized modality features | MIRFlickr I→T · mAP | 0.7605 -> 0.7278 | Hurt | Asymmetric Cross-Modal Hashing Based on Formal Concept Analysis (AAAI 2025 · 3 citations) |
| add KLM loss | Wikipedia, NUS-WIDE, XMedia, and XMediaNet · MAP | 21.6 -> 68.6; 12.2 -> 69.5; 13.4 -> 75.0; 0.9 -> 51.0 | Helped | SHE: Streaming-media Hashing Retrieval (ICML 2025) |
| Vary the number of similar concepts k | ImageNet training · training loss | Mixed | One for All: A Universal Generator for Concept Unlearnability via Multi-Modal Alignment (ICML 2024) |
What reviewers will expect
Datasets and baselines named in the 11 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- NUS-WIDE · 7 of 11 papers
- MIRFLICKR-25K · 4 of 11 papers
- MS-COCO · 4 of 11 papers
- IAPR TC-12 · 3 of 11 papers
- MIRFlickr · 2 of 11 papers
Baselines
- DSPH · 3 of 11 papers
- DJSRH · 3 of 11 papers
- JDSH · 3 of 11 papers
- DCMH · 3 of 11 papers
- DNpH · 3 of 11 papers
- DHaPH · 3 of 11 papers
- DAEH · 3 of 11 papers
- CIRH · 3 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 30 papers; from OpenAlex and the papers' affiliations).
- Central South University · 5 papers
- Beijing University of Posts and Telecommunications · 3 papers
- Sichuan University · 2 papers
- Hebei Normal University · 2 papers
- Communication University of China · 2 papers
- Information Support Force Engineering University · 2 papers
- Xiamen University · 2 papers
Use it
Download BibTeX of 36 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
- Unimodal-to-multimodal representation transfer (145 papers)
- Probabilistic embeddings for cross-modal retrieval (108 papers)
- Contrastive vision-language alignment (104 papers)
Elsewhere in the topic
- Multimodal fusion with missing modalities (208 papers)
- Multi-modality image fusion (194 papers)
- Medical multimodal fusion (166 papers)
- Multimodal sentiment analysis (106 papers)
- Audio-visual speech separation (76 papers)
- Audio-visual source localization (67 papers)
- Self-supervised audio-visual representation learning (44 papers)
- Multi-modal object re-identification (40 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).