Research map / Detection and segmentation / Open-set and few-shot detection
Human-object interaction detection
Detect human-object interactions with cross-modal, pose-aware, and knowledge-guided models.
66 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Open-set and few-shot detection, keyword cluster “human · interaction · hoi”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 5 of 6 tracked benchmarks.
16 papers in 2023–24 and 17 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 2.7% to 3.1%.
Open questions the data points to
- HICO-DET Default Full (mAP): best went from 46.01 (up to 2024) to 47.53 in 2025 — HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations)
- HICO-DET Default Non-Rare (mAP): best went from 45.8 (up to 2024) to 47.74 in 2025 — HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations)
- HICO-DET Known Object Full (mAP): best went from 40.61 (up to 2024) to 51.24 in 2025 — HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations)
- HICO-DET Known Object Rare (mAP): best went from 38.82 (up to 2024) to 50.78 in 2025 — HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations)
- HICO-DET Known Object Non-Rare (mAP): best went from 41.15 (up to 2024) to 51.38 in 2025 — HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations)
What looks settled
- HICO-DET Default Rare (mAP): best up to 2024 46.74, best in 2025–26 46.81 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
- 2016 · Progressively Parsing Interactional Objects for Fine Grained Action Detection (CVPR 2016 · 71 citations)
- 2018 · Detecting and Recognizing Human-Object Interactions (CVPR 2018 · 611 citations)
- 2019 · Transferable Interactiveness Knowledge for Human-Object Interaction Detection (CVPR 2019 · 254 citations)
- 2020 · PPDM: Parallel Point Detection and Matching for Real-Time Human-Object Interaction Detection (CVPR 2020 · 297 citations)
- 2021 · HOTR: End-to-End Human-Object Interaction Detection With Transformers (CVPR 2021 · 291 citations)
- 2022 · GEN-VLKT: Simplify Association and Enhance Interaction Understanding for HOI Detection (CVPR 2022 · 155 citations)
- 2023 · ViPLO: Vision Transformer Based Pose-Conditioned Self-Loop Graph for Human-Object Interaction Detection (CVPR 2023 · 69 citations)
- 2024 · Bilateral Adaptation for Human-Object Interaction Detection with Occlusion-Robustness (CVPR 2024 · 16 citations)
- 2025 · Locality-Aware Zero-Shot Human-Object Interaction Detection (CVPR 2025 · 6 citations)
- 2026 · QueryCraft: Transformer-Guided Query Initialization for Enhanced Human-Object Interaction Detection (AAAI 2026 · 3 citations)
Earliest papers here
- Progressively Parsing Interactional Objects for Fine Grained Action Detection (CVPR 2016 · 71 citations)
- Detecting and Recognizing Human-Object Interactions (CVPR 2018 · 611 citations)
- Transferable Interactiveness Knowledge for Human-Object Interaction Detection (CVPR 2019 · 254 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.
HICO-DET Default Rare · mAP (higher is better)
13 papers, 2019–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 13.42 | Transferable Interactiveness Knowledge for Human-Object Interaction Detection (CVPR 2019 · 254 citations) |
| 2021 | 23.67 | Detecting Human-Object Interaction via Fabricated Compositional Learning (CVPR 2021 · 113 citations) |
| 2022 | 30.3 | Interactiveness Field in Human-Object Interactions (CVPR 2022 · 45 citations) |
| 2023 | 39.91 | Detecting Any Human-Object Interaction Relationship: Universal HOI Detector with Spatial Prompt Learning on Foundation Models (NeurIPS 2023 · 10 citations) |
| 2024 | 46.74 | Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · 12 citations) |
| 2025 | 46.81 | HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations) |
HICO-DET Default Full · mAP (higher is better)
12 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 17.22 | Transferable Interactiveness Knowledge for Human-Object Interaction Detection (CVPR 2019 · 254 citations) |
| 2021 | 29.12 | Detecting Human-Object Interaction via Fabricated Compositional Learning (CVPR 2021 · 113 citations) |
| 2022 | 33.51 | Interactiveness Field in Human-Object Interactions (CVPR 2022 · 45 citations) |
| 2023 | 40.95 | Detecting Any Human-Object Interaction Relationship: Universal HOI Detector with Spatial Prompt Learning on Foundation Models (NeurIPS 2023 · 10 citations) |
| 2024 | 46.01 | Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · 12 citations) |
| 2025 | 47.53 | HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations) |
HICO-DET Default Non-Rare · mAP (higher is better)
12 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 18.11 | Transferable Interactiveness Knowledge for Human-Object Interaction Detection (CVPR 2019 · 254 citations) |
| 2021 | 26.28 | Affordance Transfer Learning for Human-Object Interaction Detection (CVPR 2021 · 117 citations) |
| 2022 | 34.46 | Interactiveness Field in Human-Object Interactions (CVPR 2022 · 45 citations) |
| 2023 | 37.75 | ViPLO: Vision Transformer Based Pose-Conditioned Self-Loop Graph for Human-Object Interaction Detection (CVPR 2023 · 69 citations) |
| 2024 | 45.8 | Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · 12 citations) |
| 2025 | 47.74 | HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations) |
HICO-DET Known Object Full · mAP (higher is better)
9 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 19.17 | Transferable Interactiveness Knowledge for Human-Object Interaction Detection (CVPR 2019 · 254 citations) |
| 2021 | 26.8 | Detecting Human-Object Interaction via Fabricated Compositional Learning (CVPR 2021 · 113 citations) |
| 2022 | 36.28 | Interactiveness Field in Human-Object Interactions (CVPR 2022 · 45 citations) |
| 2023 | 40.61 | ViPLO: Vision Transformer Based Pose-Conditioned Self-Loop Graph for Human-Object Interaction Detection (CVPR 2023 · 69 citations) |
| 2025 | 51.24 | HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations) |
HICO-DET Known Object Rare · mAP (higher is better)
7 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 21.61 | Detecting Human-Object Interaction via Fabricated Compositional Learning (CVPR 2021 · 113 citations) |
| 2022 | 33.16 | Interactiveness Field in Human-Object Interactions (CVPR 2022 · 45 citations) |
| 2023 | 38.82 | ViPLO: Vision Transformer Based Pose-Conditioned Self-Loop Graph for Human-Object Interaction Detection (CVPR 2023 · 69 citations) |
| 2025 | 50.78 | HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations) |
HICO-DET Known Object Non-Rare · mAP (higher is better)
6 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 26.92 | DIRV: Dense Interaction Region Voting for End-to-End Human-Object Interaction Detection (AAAI 2021 · 51 citations) |
| 2022 | 37.21 | Interactiveness Field in Human-Object Interactions (CVPR 2022 · 45 citations) |
| 2023 | 41.15 | ViPLO: Vision Transformer Based Pose-Conditioned Self-Loop Graph for Human-Object Interaction Detection (CVPR 2023 · 69 citations) |
| 2025 | 51.38 | HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · 4 citations) |
What has been tried
58 papers here report 712 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 3%, hurt 19%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add spatial adaptation to semantic adaptation | HICO-DET · mAP | 36.33 -> 39.34 | Helped | Bilateral Adaptation for Human-Object Interaction Detection with Occlusion-Robustness (CVPR 2024 · 16 citations) |
| add self-triplet aggregation (STA) on top of KD | HICO-DET · mAP Full | +0.78 (35.63 to 36.41) | Helped | Exploring Self- and Cross-Triplet Correlations for Human-Object Interaction Detection (AAAI 2024 · 13 citations) |
| Add the syntactic interaction bank (SIB) | HICO-DET Default · mAP (Full) | 35.71 -> 39.34 | Helped | Discovering Syntactic Interaction Clues for Human-Object Interaction Detection (CVPR 2024 · 13 citations) |
| Add Haa500 action data | HICO-DET DT Full · mAP | 31.98 -> 33.32 | Helped | Disentangled Pre-training for Human-Object Interaction Detection (CVPR 2024 · 13 citations) |
| replace CLIP-based VLKT with BLIP-based VLKT | HICO-DET Rare · mAP | 29.25 -> 30.49 | Helped | Dual-Prior Augmented Decoding Network for Long Tail Distribution in HOI Detection (AAAI 2024 · 12 citations) |
| add hybrid learning with decaying λ | HICO-DET Default Full · mAP | 33.44 -> 34.42 | Helped | Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · 12 citations) |
| add spatial priors | Ego4D · AP | 35.9 -> 36.1 | Helped | Active Object Detection with Knowledge Aggregation and Distillation from Large Models (CVPR 2024 · 9 citations) |
| add verb contrastive learning | VG-HOI · mAP | 8.85 -> 9.20 | Helped | Toward Open-Set Human Object Interaction Detection (AAAI 2024 · 7 citations) |
| Add Locality Adapter to CLIP representations | HICO-DET UV · mAP | 24.88 -> 27.71 | Helped | Locality-Aware Zero-Shot Human-Object Interaction Detection (CVPR 2025 · 6 citations) |
| add the context branch to the base model | HICO-DET Default full · mAP | 34.69 -> 35.81 | Helped | ContextHOI: Spatial Context Learning for Human-Object Interaction Detection (AAAI 2025 · 5 citations) |
| Add OI (object inpainting) to the baseline | HOT-Annotated · SC-Acc. | 40.5 -> 42.4 (+7%) | Helped | Precision-Enhanced Human-Object Contact Detection via Depth-Aware Perspective Interaction and Object Texture Restoration (AAAI 2025 · 4 citations) |
| Add object key points to the baseline | FullBodyManipulation test · FID | 9.972 -> 7.514 | Helped | Guiding Human-Object Interactions with Rich Geometry and Relations (CVPR 2025 · 4 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 |
|---|---|---|---|---|
| HOICLIP | HICO-DET UV Unseen | mAP | 24.3 – 26.23 | 3 papers, e.g. AAAI 2024 and CVPR 2025 |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- V-COCO · 7 of 9 papers
- HICO-DET · 6 of 9 papers
- MS-COCO · 3 of 9 papers
- NVI · 2 of 9 papers
Baselines
- GEN-VLKT · 5 of 9 papers
- QPIC · 4 of 9 papers
- HOICLIP · 4 of 9 papers
- ADA-CM · 4 of 9 papers
- PViC · 4 of 9 papers
- UPT · 3 of 9 papers
- RLIPv2 · 3 of 9 papers
- CDN · 3 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 33 papers; from OpenAlex and the papers' affiliations).
- Peking University · 3 papers
- Zhejiang University · 3 papers
- Pohang University of Science and Technology · 3 papers
- Tencent · 2 papers
- Xiamen University · 2 papers
- Alibaba Group · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- South China University of Technology · 2 papers
Use it
Download BibTeX of 40 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
- Unsupervised object and part discovery (114 papers)
- Visual relationship and scene graph detection (63 papers)
- Hand-object and articulated interaction (52 papers)
Elsewhere in the topic
- Object-level representation learning (219 papers)
- Interactive and referring segmentation (161 papers)
- Open-set and unknown object detection (160 papers)
- Amodal and open-world instance segmentation (137 papers)
- Industrial anomaly detection (117 papers)
- Domain-adaptive object detection (103 papers)
- Open-vocabulary object detection (96 papers)
- Open-vocabulary semantic segmentation (93 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).