Research map / 3D vision / Human pose estimation
Learning 3D Human Pose from Structure and Motion
Rishabh Dabral, Anurag Mundhada, Uday Kusupati, Safeer Afaque, Abhishek Sharma, Arjun Jain
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 geometry loss (ablation) | baseline (fully supervised) · Human3.6M (Protocol 1) | MPJPE: 58.50 -> 58.45 | No clear effect |
| add illegal-angle loss (ablation) | baseline + geometry loss · Human3.6M (Protocol 1) | MPJPE: 58.45 -> 56.20 | Improved |
| add symmetry loss (ablation) | baseline + geometry + illegal-angle · Human3.6M (Protocol 1) | MPJPE: 56.20 -> 55.51 | Improved |
| add geometry loss to Zhou et al. weak supervision (ablation) | Zhou et al. · Human3.6M (Protocol 1) | MPJPE: 65.69 -> 64.90 | Improved |
| add real-time TP-Net temporal model (ablation) | SAP-Net · Human3.6M (Protocol 1) | MPJPE: 55.51 -> 52.10 | Improved |
| make TP-Net bi-directional (semi-online) (ablation) | TP-Net · Human3.6M (Protocol 1) | MPJPE: 52.10 -> 51.10 | Improved |
| replace LSTM with fully-connected TP-Net (ablation) | temporal model · Human3.6M (Protocol 1) | MPJPE: 54.05 -> 52.1 | Improved |
| increase TP-Net input frames from 4 to 20 (ablation) | TP-Net · Human3.6M (Protocol 1) | MPJPE: 53.0 -> 52.1 | Improved |
| increase Bi-LSTM input frames from 4 to 20 (ablation) | Bi-LSTM · Human3.6M (Protocol 1) | MPJPE: 53.86 -> 53.65 | Improved |
| add TP-Net to SAP-Net (ablation) | SAP-Net · MPI-INF-3DHP | bone-length standard deviation: reduces the standard deviation of bone-length across the frames by 28.7% | Improved |
| add symmetry loss and TP-Net vs bone-length-ratio prior only (ablation) | SAP-Net · MPI-INF-3DHP | illegal non-torso joint angles: 1.4% -> 0.8% | Improved |
Closest papers by meaning
- Kinematic-Structure-Preserved Representation for Unsupervised 3D Human Pose Estimation (AAAI 2020 · similarity 0.92)
- Towards 3D Human Pose Estimation in the Wild: A Weakly-Supervised Approach (ICCV 2017 · similarity 0.92)
- Weakly-Supervised 3D Human Pose Learning via Multi-View Images in the Wild (CVPR 2020 · similarity 0.92)
- Self-Supervised 3D Human Pose Estimation via Part Guided Novel Image Synthesis (CVPR 2020 · similarity 0.92)
- Exploiting temporal information for 3D human pose estimation (ECCV 2018 · similarity 0.92)
- Weakly-Supervised Discovery of Geometry-Aware Representation for 3D Human Pose Estimation (CVPR 2019 · similarity 0.91)
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.
