Research map / 3D vision / Human pose estimation
Temporal video pose estimation
Estimates human pose in video using temporal difference learning and sparse labels.
76 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 Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 2 of 2 tracked benchmarks.
10 papers in 2023–24 and 18 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.7% to 6.0%.
Open questions the data points to
- PoseTrack2017 validation (mAP): best went from 85.9 (up to 2024) to 87.6 in 2025 — Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · 3 citations)
- PoseTrack2018 validation (mAP): best went from 83.6 (up to 2024) to 84.6 in 2025 — Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · 3 citations)
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 · Personalizing Human Video Pose Estimation (CVPR 2016 · 98 citations)
- 2017 · ArtTrack: Articulated Multi-Person Tracking in the Wild (CVPR 2017 · 255 citations)
- 2018 · PoseTrack: A Benchmark for Human Pose Estimation and Tracking (CVPR 2018 · 516 citations)
- 2019 · Learning Individual Styles of Conversational Gesture (CVPR 2019 · 346 citations)
- 2020 · Combining Detection and Tracking for Human Pose Estimation in Videos (CVPR 2020 · 134 citations)
- 2021 · Beyond Static Features for Temporally Consistent 3D Human Pose and Shape From a Video (CVPR 2021 · 257 citations)
- 2022 · HumanNeRF: Free-Viewpoint Rendering of Moving People From Monocular Video (CVPR 2022 · 450 citations)
- 2023 · Global-to-Local Modeling for Video-Based 3D Human Pose and Shape Estimation (CVPR 2023 · 79 citations)
- 2024 · Hourglass Tokenizer for Efficient Transformer-Based 3D Human Pose Estimation (CVPR 2024 · 67 citations)
- 2025 · SpatioTemporal Learning for Human Pose Estimation in Sparsely-Labeled Videos (AAAI 2025 · 4 citations)
- 2026 · End-to-End Multi-Person Pose Estimation with Pose-Aware Video Transformer (AAAI 2026 · 3 citations)
Earliest papers here
- Personalizing Human Video Pose Estimation (CVPR 2016 · 98 citations)
- Recognizing Activities of Daily Living With a Wrist-Mounted Camera (CVPR 2016 · 58 citations)
- ArtTrack: Articulated Multi-Person Tracking in the Wild (CVPR 2017 · 255 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.
PoseTrack2017 validation · mAP (higher is better)
4 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 72.6 | Efficient Online Multi-Person 2D Pose Tracking With Recurrent Spatio-Temporal Affinity Fields (CVPR 2019 · 137 citations) |
| 2022 | 84.8 | Temporal Feature Alignment and Mutual Information Maximization for Video-Based Human Pose Estimation (CVPR 2022 · 79 citations) |
| 2023 | 85.9 | Mutual Information-Based Temporal Difference Learning for Human Pose Estimation in Video (CVPR 2023 · 28 citations) |
| 2025 | 87.6 | Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · 3 citations) |
PoseTrack2018 validation · mAP (higher is better)
4 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 70.4 | Efficient Online Multi-Person 2D Pose Tracking With Recurrent Spatio-Temporal Affinity Fields (CVPR 2019 · 137 citations) |
| 2022 | 82.2 | Temporal Feature Alignment and Mutual Information Maximization for Video-Based Human Pose Estimation (CVPR 2022 · 79 citations) |
| 2023 | 83.6 | Mutual Information-Based Temporal Difference Learning for Human Pose Estimation in Video (CVPR 2023 · 28 citations) |
| 2025 | 84.6 | Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · 3 citations) |
What has been tried
67 papers here report 714 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 f=9 representative tokens instead of the baseline configuration | Human3.6M, seq2seq · MPJPE (mm) | 40.9 -> 43.5 | Hurt | Hourglass Tokenizer for Efficient Transformer-Based 3D Human Pose Estimation (CVPR 2024 · 67 citations) |
| Use two source images instead of one | VoxCeleb test set, self-reenactment · PSNR | 23.67 -> 25.00 | Helped | FSRT: Facial Scene Representation Transformer for Face Reenactment from Factorized Appearance Head-pose and Facial Expression Features (CVPR 2024 · 24 citations) |
| Add JFD and STD to represent and aggregate joint-level features | PoseTrack2017 val. · mAP | 73.8 -> 78.6 | Helped | Video-Based Human Pose Regression via Decoupled Space-Time Aggregation (CVPR 2024 · 15 citations) |
| add hierarchical cross-entropy loss to SynNet | IFID · Accin@1 | 20.56% -> 31.78% | Helped | Advancing Video Synchronization with Fractional Frame Analysis: Introducing a Novel Dataset and Model (AAAI 2024 · 5 citations) |
| Add Temporal Feature Fusion (TFF) | PoseTrack2017 Val Set · mAP | 86.8 -> 88.4; (\uparrow 1.6) | Helped | SpatioTemporal Learning for Human Pose Estimation in Sparsely-Labeled Videos (AAAI 2025 · 4 citations) |
| Train and infer using sequences of 2 frames instead of the full sequence setup | Sintel · ATE | 0.078 -> 0.134 | Hurt | AnyCam: Learning to Recover Camera Poses and Intrinsics from Casual Videos (CVPR 2025 · 4 citations) |
| Add optical-flow filtering | Panda-Test · precision and recall | Helped | Dynamic Camera Poses and Where to Find Them (CVPR 2025 · 4 citations) | |
| add HKME to the ViT pose baseline | PoseTrack2017 validation · mAP | 80.2 -> 85.3; 5.1 mAP | Helped | Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · 3 citations) |
| Add 3DPW training data to BEDLAM | 3DPW test · MPJPE | 82.18 -> 75.86 | Helped | Motions as Queries: One-Stage Multi-Person Holistic Human Motion Capture (CVPR 2025 · 3 citations) |
| add masked token reconstruction task | PoseTrack2017 val · Mean | 87.0 -> 87.5 | Helped | Causal-Inspired Multitask Learning for Video-Based Human Pose Estimation (AAAI 2025 · 2 citations) |
| Remove uncertainty guidance | AMASS test, Ts=20 · MPJPE | 6.05 -> 5.51 | Hurt | Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data (NeurIPS 2024 · 2 citations) |
| Increase context length from 3 to 15 frames | Nymeria validation, 2 seconds ahead · LPIPS | 0.304^{0.002} -> 0.303^{0.001} | Helped | Whole-Body Conditioned Egocentric Video Prediction (NeurIPS 2025 · 1 citation) |
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
- PoseTrack2017 · 3 of 11 papers
- PoseTrack2018 · 3 of 11 papers
- PoseTrack2021 · 3 of 11 papers
- COCO · 2 of 11 papers
Baselines
- PoseWarper · 3 of 11 papers
- DCPose · 3 of 11 papers
- FAMI-Pose · 3 of 11 papers
- DSTA · 3 of 11 papers
- DUSt3R · 3 of 11 papers
- HANet · 2 of 11 papers
- M-HANet · 2 of 11 papers
- TDMI · 2 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 28 papers; from OpenAlex and the papers' affiliations).
- Jilin University · 4 papers
- Zhejiang Gongshang University · 4 papers
- Zhejiang University · 3 papers
- Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security · 3 papers
- Tsinghua University · 2 papers
- National University of Singapore · 2 papers
- Meta · 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
- Monocular 2D-to-3D pose lifting (90 papers)
- Transformer-based multi-view 3D pose (83 papers)
- Multi-person pose estimation (51 papers)
- Heatmap and coordinate pose encoding (50 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).