Research map / 3D vision / Human pose estimation
3D Human Pose Machines with Self-supervised Learning
Keze Wang, Liang Lin, Chenhan Jiang, Chen Qian, Pengxu Wei
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 self-supervised correction in training (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 80.15 -> 67.63 | Improved |
| add self-supervised correction at test time (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 67.63 -> 63.67 | Improved |
| add 3D-to-2D projection instead of direct 2D and 3D multi-task regression (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 86.80 -> 63.67 | Improved |
| add temporal LSTM dependency in training (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 90.83 -> 63.67 | Improved |
| use sequence input at test time (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 67.70 -> 63.67 | Improved |
| add external 2D pose data from MPII (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 111.3 -> 63.67 | Improved |
| use 2D ground truth for correction (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 63.67 -> 59.41 | Improved |
| use hourglass features for 2D pose network (ablation) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 63.67 -> 62.85 | Improved |
| replace multi-stage refinement with 3D-to-2D projector (vs conference version) | 3D pose machine · Human3.6M Protocol #1 | 3D pose error (mm): 73.10 -> 63.67 | Improved |
| compare to best prior method under per-action training | 3D pose machine · Human3.6M Protocol #1 (per action) | 3D pose error (mm): 69.73 -> 63.79 | Improved |
| compare to best prior method | 3D pose machine · Human3.6M Protocol #2 | 3D pose error (mm): 79.6 -> 63.74 | Improved |
| compare to best prior method | 3D pose machine · Human3.6M Protocol #3 | 3D pose error (mm): 70.70 -> 54.14 | Improved |
Closest papers by meaning
- Geometry-Driven Self-Supervised Method for 3D Human Pose Estimation (AAAI 2020 · similarity 0.92)
- Self-Supervised 3D Human Pose Estimation via Part Guided Novel Image Synthesis (CVPR 2020 · similarity 0.92)
- SelfPose3d: Self-Supervised Multi-Person Multi-View 3d Pose Estimation (CVPR 2024 · similarity 0.91)
- Self-Supervised 3D Human Mesh Recovery from a Single Image with Uncertainty-Aware Learning (AAAI 2024 · similarity 0.89)
- PoseIRM: Enhance 3D Human Pose Estimation on Unseen Camera Settings via Invariant Risk Minimization (CVPR 2024 · similarity 0.89)
- PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space (NeurIPS 2025 · similarity 0.89)
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.
