Research map / 3D vision / Human pose estimation
Multi-person pose estimation
Detects and assigns keypoints for many people using top-down, bottom-up or single-stage approaches.
51 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Human pose estimation, keyword cluster “human pose · pose estimation · 3d human”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
12 papers in 2023–24 and 7 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 4.4% to 2.3%.
Open questions the data points to
- Papers report DEKR on OCHuman Test (AP) anywhere from 38.2 to 52.2: comparisons against it depend on whose number you take.
- Papers report Baseline on COCO val2017 (AP) anywhere from 60 to 74.3: comparisons against it depend on whose number you take.
No longer reported by papers here: CrowdPose test (AP, last in 2024); COCO test-dev (AP, last in 2024); COCO val2017 (AP, last in 2024); MuPoTS-3D (PCK_abs, last in 2022).
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 · DeepCut: Joint Subset Partition and Labeling for Multi Person Pose Estimation (CVPR 2016 · 1,103 citations)
- 2017 · Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields (CVPR 2017 · 7,514 citations)
- 2018 · Attention-Aware Compositional Network for Person Re-Identification (CVPR 2018 · 493 citations)
- 2019 · CrowdPose: Efficient Crowded Scenes Pose Estimation and a New Benchmark (CVPR 2019 · 623 citations)
- 2020 · UniPose: Unified Human Pose Estimation in Single Images and Videos (CVPR 2020 · 191 citations)
- 2021 · TesseTrack: End-to-End Learnable Multi-Person Articulated 3D Pose Tracking (CVPR 2021 · 128 citations)
- 2022 · End-to-End Multi-Person Pose Estimation With Transformers (CVPR 2022 · 164 citations)
- 2023 · 3D Human Pose Estimation With Spatio-Temporal Criss-Cross Attention (CVPR 2023 · 218 citations)
- 2024 · RTMO: Towards High-Performance One-Stage Real-Time Multi-Person Pose Estimation (CVPR 2024 · 97 citations)
- 2025 · DynPose: Largely Improving the Efficiency of Human Pose Estimation by a Simple Dynamic Framework (CVPR 2025 · 3 citations)
- 2026 · A Cross-view Fusion Framework for Robust 6-DoF Grasp Pose Estimation (CVPR 2026 · 0 citations)
Earliest papers here
- DeepCut: Joint Subset Partition and Labeling for Multi Person Pose Estimation (CVPR 2016 · 1,103 citations)
- Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields (CVPR 2017 · 7,514 citations)
- Multi-Context Attention for Human Pose Estimation (CVPR 2017 · 739 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.
CrowdPose test · AP (higher is better)
6 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 74.3 | RSGNet: Relation based Skeleton Graph Network for Crowded Scenes Pose Estimation (AAAI 2021 · 15 citations) |
COCO test-dev · AP (higher is better)
5 papers, 2017–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 61.8 | Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields (CVPR 2017 · 7,514 citations) |
| 2021 | 76 | RSGNet: Relation based Skeleton Graph Network for Crowded Scenes Pose Estimation (AAAI 2021 · 15 citations) |
COCO val2017 · AP (higher is better)
5 papers, 2022–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 76.4 | Adaptive Hypergraph Neural Network for Multi-Person Pose Estimation (AAAI 2022 · 12 citations) |
MuPoTS-3D · PCK_abs (higher is better)
4 papers, 2021–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 48 | Monocular 3D Multi-Person Pose Estimation by Integrating Top-Down and Bottom-Up Networks (CVPR 2021 · 49 citations) |
What has been tried
46 papers here report 501 measured changes (a component added, removed or swapped, compared with a baseline): helped 78%, no clear effect 4%, hurt 16%, 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 |
|---|---|---|---|---|
| use coordinate classification with DCC and MLE instead of regression | COCO val2017 · AP | 65.6 -> 67.6 | Helped | RTMO: Towards High-Performance One-Stage Real-Time Multi-Person Pose Estimation (CVPR 2024 · 97 citations) |
| Add a Refine decoder | All test samples · MPJPE | 116.1 -> 107.2 | Helped | Person-in-WiFi 3D: End-to-End Multi-Person 3D Pose Estimation with Wi-Fi (CVPR 2024 · 37 citations) |
| use ground-truth 2d poses instead of pseudo 2d poses | Panoptic · MPJPE | 24.5 -> 19.9 | Helped | SelfPose3d: Self-Supervised Multi-Person Multi-View 3d Pose Estimation (CVPR 2024 · 22 citations) |
| set signal scale to 5.0 instead of 1.0 | CrowdPose test · AP | 72.4 -> 72.7 | Helped | DiffusionRegPose: Enhancing Multi-Person Pose Estimation using a Diffusion-Based End-to-End Regression Approach (CVPR 2024 · 9 citations) |
| Use router-controlled routing instead of random routing | COCO · AP | Helped | DynPose: Largely Improving the Efficiency of Human Pose Estimation by a Simple Dynamic Framework (CVPR 2025 · 3 citations) | |
| Add recurrent GRU update to the hidden state | PoseTrack · 2D@0.05 | 73.0 -> 74.9 | Helped | CoMotion: Concurrent Multi-person 3D Motion (ICLR 2025 · 0 citations) |
| pretrain on pose datasets and train on PVCP instead of PVCP training alone | PVCP testset (All), 2D Det input · MPVE (mm) | 315.64 -> 282.50 | Helped | Pedestrian-Centric 3D Pre-collision Pose and Shape Estimation from Dashcam Perspective (NeurIPS 2024 · 0 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 |
|---|---|---|---|---|
| DEKR | OCHuman Test | AP | 38.2 – 52.2 | 2 papers, e.g. CVPR 2022 and NeurIPS 2021 |
| Baseline | COCO val2017 | AP | 60 – 74.3 | 2 papers, e.g. CVPR 2022 and AAAI 2022 |
| Kocabas et al. | 3DPW occlusion subset | PA-MPJPE | 72.2 – 82.5 | 2 papers, e.g. CVPR 2021 and AAAI 2021 |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MPII · 3 of 9 papers
- JHMDB · 2 of 9 papers
- COCO · 2 of 9 papers
- AMASS · 2 of 9 papers
Baselines
- ED-Pose · 3 of 9 papers
- PETR · 3 of 9 papers
- DEKR · 3 of 9 papers
- SimpleBaseline · 2 of 9 papers
- RTMPose · 2 of 9 papers
- CID · 2 of 9 papers
- MotionBERT · 2 of 9 papers
- PyMAF · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 19 papers; from OpenAlex and the papers' affiliations).
- Stanford University · 2 papers
- Alibaba Group · 2 papers
Use it
Download BibTeX of 32 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
- Monocular 2D-to-3D pose lifting (90 papers)
- Transformer-based multi-view 3D pose (83 papers)
- Heatmap and coordinate pose encoding (50 papers)
- Temporal video pose estimation (76 papers)
Elsewhere in the topic
- Physics and sensor-based motion capture (125 papers)
- Learned camera pose estimation (96 papers)
- Category-level object pose estimation (92 papers)
- Depth-based hand pose regression (84 papers)
- Human shape and pose regression (80 papers)
- Instance-level 6D object pose (74 papers)
- Human mesh recovery from images (58 papers)
- Category-level pose estimation (54 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).