Research map / 3D vision / Human pose estimation
Monocular 2D-to-3D pose lifting
Lifts single-image or video 2D keypoints to 3D with multi-hypothesis, weak-supervision and adaptation methods.
90 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
Still moving The best reported result is still improving on 3 of 6 tracked benchmarks.
10 papers in 2023–24 and 20 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.7% to 6.7%.
Open questions the data points to
- MPI-INF-3DHP (AUC): best went from 77 (up to 2024) to 78.7 in 2025 — CHAMP: Conformalized 3D Human Multi-Hypothesis Pose Estimators (ICLR 2025 · 0 citations)
- 3DPW (PA-MPJPE): best went from 46.4 (up to 2024) to 44.3 in 2025 — Semantics-aware Test-time Adaptation for 3D Human Pose Estimation (ICML 2025 · 0 citations)
- 3DPW (MPJPE): best went from 73 (up to 2024) to 71.2 in 2025 — Semantics-aware Test-time Adaptation for 3D Human Pose Estimation (ICML 2025 · 0 citations)
- Papers report Baseline on Human3.6M Protocol #1 (MPJPE (mm)) anywhere from 51.8 to 64.8: comparisons against it depend on whose number you take.
No longer reported by papers here: Human3.6M Protocol #1 (MPJPE (mm), last in 2023).
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 · Sparseness Meets Deepness: 3D Human Pose Estimation From Monocular Video (CVPR 2016 · 440 citations)
- 2017 · Coarse-To-Fine Volumetric Prediction for Single-Image 3D Human Pose (CVPR 2017 · 947 citations)
- 2018 · 2D/3D Pose Estimation and Action Recognition Using Multitask Deep Learning (CVPR 2018 · 594 citations)
- 2019 · Self-Supervised Learning of 3D Human Pose Using Multi-View Geometry (CVPR 2019 · 316 citations)
- 2020 · Attention Mechanism Exploits Temporal Contexts: Real-Time 3D Human Pose Reconstruction (CVPR 2020 · 275 citations)
- 2021 · Graph Stacked Hourglass Networks for 3D Human Pose Estimation (CVPR 2021 · 298 citations)
- 2022 · Uncertainty-Aware Adaptation for Self-Supervised 3D Human Pose Estimation (CVPR 2022 · 41 citations)
- 2023 · Unified Pose Sequence Modeling (CVPR 2023 · 43 citations)
- 2024 · ChatPose: Chatting about 3D Human Pose (CVPR 2024 · 54 citations)
- 2025 · TCPFormer: Learning Temporal Correlation with Implicit Pose Proxy for 3D Human Pose Estimation (AAAI 2025 · 35 citations)
- 2026 · Robust Long-Term Test-Time Adaptation for 3D Human Pose Estimation Through Motion Discretization (AAAI 2026 · 1 citation)
Earliest papers here
- Sparseness Meets Deepness: 3D Human Pose Estimation From Monocular Video (CVPR 2016 · 440 citations)
- Structured Feature Learning for Pose Estimation (CVPR 2016 · 288 citations)
- Direct Prediction of 3D Body Poses From Motion Compensated Sequences (CVPR 2016 · 244 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.
MPI-INF-3DHP · AUC (higher is better)
8 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 54.8 | RepNet: Weakly Supervised Training of an Adversarial Reprojection Network for 3D Human Pose Estimation (CVPR 2019 · 278 citations) |
| 2023 | 76.3 | A Single 2D Pose with Context is Worth Hundreds for 3D Human Pose Estimation (NeurIPS 2023 · 8 citations) |
| 2024 | 77 | ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation (NeurIPS 2024 · 3 citations) |
| 2025 | 78.7 | CHAMP: Conformalized 3D Human Multi-Hypothesis Pose Estimators (ICLR 2025 · 0 citations) |
Human3.6M · MPJPE (lower is better)
7 papers, 2019–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 60.56 | Self-Supervised Learning of 3D Human Pose Using Multi-View Geometry (CVPR 2019 · 316 citations) |
| 2021 | 53.6 | Reconstructing 3D Human Pose by Watching Humans in the Mirror (CVPR 2021 · 59 citations) |
| 2024 | 39.1 | ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation (NeurIPS 2024 · 3 citations) |
MPI-INF-3DHP · PCK (higher is better)
7 papers, 2019–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 64.7 | Self-Supervised Learning of 3D Human Pose Using Multi-View Geometry (CVPR 2019 · 316 citations) |
| 2020 | 80.2 | Kinematic-Structure-Preserved Representation for Unsupervised 3D Human Pose Estimation (AAAI 2020 · 39 citations) |
| 2023 | 98.2 | A Single 2D Pose with Context is Worth Hundreds for 3D Human Pose Estimation (NeurIPS 2023 · 8 citations) |
| 2024 | 98.4 | ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation (NeurIPS 2024 · 3 citations) |
3DPW · PA-MPJPE (lower is better)
6 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 79.1 | Aligning Silhouette Topology for Self-Adaptive 3D Human Pose Recovery (NeurIPS 2021 · 6 citations) |
| 2022 | 64.1 | ElePose: Unsupervised 3D Human Pose Estimation by Predicting Camera Elevation and Learning Normalizing Flows on 2D Poses (CVPR 2022 · 40 citations) |
| 2024 | 46.4 | PoseGen: Learning to Generate 3D Human Pose Dataset with NeRF (AAAI 2024 · 3 citations) |
| 2025 | 44.3 | Semantics-aware Test-time Adaptation for 3D Human Pose Estimation (ICML 2025 · 0 citations) |
Human3.6M Protocol #1 · MPJPE (mm) (lower is better)
6 papers, 2018–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 58.6 | 3D Human Pose Estimation in the Wild by Adversarial Learning (CVPR 2018 · 425 citations) |
| 2019 | 46.3 | Weakly-Supervised Discovery of Geometry-Aware Representation for 3D Human Pose Estimation (CVPR 2019 · 117 citations) |
| 2020 | 45.6 | Deep Kinematics Analysis for Monocular 3D Human Pose Estimation (CVPR 2020 · 211 citations) |
| 2023 | 25.8 | Triangulation Residual Loss for Data-efficient 3D Pose Estimation (NeurIPS 2023 · 3 citations) |
3DPW · MPJPE (lower is better)
5 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 126.3 | Aligning Silhouette Topology for Self-Adaptive 3D Human Pose Recovery (NeurIPS 2021 · 6 citations) |
| 2024 | 73 | PoseGen: Learning to Generate 3D Human Pose Dataset with NeRF (AAAI 2024 · 3 citations) |
| 2025 | 71.2 | Semantics-aware Test-time Adaptation for 3D Human Pose Estimation (ICML 2025 · 0 citations) |
What has been tried
78 papers here report 830 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 4%, hurt 17%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add differential generation | 3DHP test · MPJPE | 5.9mm increase in MPJPE | Helped | A Dual-Augmentor Framework for Domain Generalization in 3D Human Pose Estimation (CVPR 2024 · 49 citations) |
| Add PUM to the proxy-only model | Human3.6M, T=243 · MPJPE | 42.2 -> 39.5 | Helped | TCPFormer: Learning Temporal Correlation with Implicit Pose Proxy for 3D Human Pose Estimation (AAAI 2025 · 35 citations) |
| add keypoint prototype learning | COCO, AIC, MPII, AP-10K, APT-36K, COCO-WholeBody · average score | 68.2 -> 70.4 | Helped | PoseBH: Prototypical Multi-Dataset Training Beyond Human Pose Estimation (CVPR 2025 · 4 citations) |
| Add novel pose generation | 3DPW · MPJPE | 96.9 -> 95.8 | Helped | PoseGen: Learning to Generate 3D Human Pose Dataset with NeRF (AAAI 2024 · 3 citations) |
| add multiple hypotheses (K=5) | Human3.6M · MPJPE | 44.6 -> 39.1 | Helped | ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation (NeurIPS 2024 · 3 citations) |
| Add multi-scale masking to the vanilla architecture | SMPL-NPT unseen · PMD (×10−4) | 4.0 -> 3.8 | Helped | Towards Robust 3D Pose Transfer with Adversarial Learning (CVPR 2024 · 2 citations) |
| Add LMM contact loss to optimization | Two-person validation examples · PA-MPJPE | 81 -> 138 | Hurt | Pose Priors from Language Models (CVPR 2025 · 1 citation) |
| add the pose SSM module to the GCN-Transformer estimator | Human3.6M, CPN · MPJPE (mm), Protocol #1 | 48.7 -> 48.1 | Helped | Sequential Joint Dependency Aware Human Pose Estimation with State Space Model (AAAI 2025 · 0 citations) |
| add semantics-aware motion-text alignment | 3DPW · MPJPE | 87.7 to 79.3 | Helped | Semantics-aware Test-time Adaptation for 3D Human Pose Estimation (ICML 2025 · 0 citations) |
| use multiple randomly masked hypotheses during training | 3DPW · MPJPE | Helped | CUPS: Improving Human Pose-Shape Estimators with Conformalized Deep Uncertainty (ICML 2025 · 0 citations) | |
| Add global fixed anchors to a model without anchors | Human3.6M full test set · MPJPE (mm) | 42.1 -> 40.8 | Helped | PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space (NeurIPS 2025 · 0 citations) |
| Add conformal filtering before mean aggregation | Human3.6M · MPJPE (Protocol #1, mm) | 39.2 -> 36.9 | Helped | CHAMP: Conformalized 3D Human Multi-Hypothesis Pose Estimators (ICLR 2025 · 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 |
|---|---|---|---|---|
| Baseline | Human3.6M Protocol #1 | MPJPE (mm) | 51.8 – 64.8 | 2 papers, e.g. CVPR 2020 and CVPR 2018 |
What reviewers will expect
Datasets and baselines named in the 10 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Human3.6M · 6 of 10 papers
- 3DPW · 5 of 10 papers
- MPI-INF-3DHP · 4 of 10 papers
- MPII · 2 of 10 papers
Baselines
- MHFormer · 3 of 10 papers
- MixSTE · 3 of 10 papers
- GraFormer · 2 of 10 papers
- P-STMO · 2 of 10 papers
- KTPFormer · 2 of 10 papers
- POCO · 2 of 10 papers
- 3DMB · 2 of 10 papers
- DiffPose · 2 of 10 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).
- Tsinghua University · 3 papers
- University of Central Florida · 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
- Transformer-based multi-view 3D pose (83 papers)
- Multi-person pose estimation (51 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).