Research map / 3D vision / Human pose estimation
Exploiting Temporal Context for 3D Human Pose Estimation in the Wild
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add the pose prior to reprojection-only bundle adjustment (ablation) | Human 3.6M pose fitting · Human 3.6M validation | MPJPE (mm): 154.3 -> 79.6 | Helped |
| Add temporal consistency for joints and camera parameters (ablation) | Human 3.6M pose fitting with E_R + E_P · Human 3.6M validation | MPJPE (mm): 79.6 -> 77.8 | Helped |
| Use reprojection error alone, without the pose prior and temporal term (ablation) | Human 3.6M pose fitting · Human 3.6M validation | MPJPE (mm): 85.8 -> 154.3 | Hurt |
| Use ground-truth 2D keypoints instead of detected keypoints (ablation) | Bundle adjustment with E_R + E_P + E_T · Human 3.6M validation | MPJPE (mm): 77.8 -> 63.3 | Helped |
| Retrain HMR on Kinetics 300K alongside original data | HMR · 3DPW | PA-MPJPE (mm): 77.2 -> 73.8 | Helped |
| Increase Kinetics training data from 300K to 3M frames | HMR retrained with original data · 3DPW | PA-MPJPE (mm): 73.8 -> 72.2 | Helped |
| Retrain HMR on Kinetics 300K alongside original data | HMR · HumanEVA | PA-MPJPE (mm): 85.7 -> 83.5 | Helped |
| Increase Kinetics training data from 300K to 3M frames | HMR retrained with original data · HumanEVA | PA-MPJPE (mm): 83.5 -> 82.1 | Helped |
| Retrain HMR on Kinetics 300K alongside original data | HMR · Ordinal Depth | accuracy (%): 82.5 -> 83.7 | Helped |
| Train on PA-aligned 3D keypoints only, rather than original losses (ablation) | HMR trained on original data · 3DPW | PA-MPJPE (mm): 77.2 -> 78.7 | Hurt |
| Remove original 2D training data and its adversarial prior (ablation) | HMR trained on original data with PA-only 3D · 3DPW | PA-MPJPE (mm): 78.7 -> 144.6 | Hurt |
| Apply the modified training procedure without adding Kinetics data (ablation) | HMR trained on original data · HumanEVA | PA-MPJPE (mm): 85.7 -> 86.2 | Hurt |
Closest papers by meaning
- Direct Prediction of 3D Body Poses From Motion Compensated Sequences (CVPR 2016 · similarity 0.91)
- Capturing the Motion of Every Joint: 3D Human Pose and Shape Estimation with Independent Tokens (ICLR 2023 · similarity 0.90)
- Beyond Static Features for Temporally Consistent 3D Human Pose and Shape From a Video (CVPR 2021 · similarity 0.90)
- MultiPly: Reconstruction of Multiple People from Monocular Video in the Wild (CVPR 2024 · similarity 0.89)
- HumanMM: Global Human Motion Recovery from Multi-shot Videos (CVPR 2025 · similarity 0.89)
- TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose Representation (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.