Research map / 3D vision / Human pose estimation
Sequential 3D Human Pose and Shape Estimation From Point Clouds
Kangkan Wang, Jin Xie, Guofeng Zhang, Lei Liu, Jian Yang
Check 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 |
|---|---|---|---|
| Predict mesh vertex coordinates directly with spatial-temporal mesh attention convolution | Sequential point-cloud body reconstruction · SURREAL | Reconstruction error (mm): 49.5 -> 18.2 | Improved |
| Predict mesh vertex coordinates directly with spatial-temporal mesh attention convolution | Sequential point-cloud body reconstruction · Human3.6M | Reconstruction error (mm): 54.3 -> 21.4 | Improved |
| Predict mesh vertex coordinates directly with spatial-temporal mesh attention convolution | Sequential point-cloud body reconstruction · DFAUST | Reconstruction error (mm): 52.2 -> 19.7 | Improved |
| Add spatial mesh attention convolution (SMAC) (ablation) | Proposed sequential mesh reconstruction · SURREAL, Human3.6M, DFAUST | Reconstruction error (mm): 45.3 -> 18.2; 49.1 -> 21.4; 47.6 -> 19.7 | Improved |
| Use learned spatial attention instead of neighborhood averaging (ablation) | Proposed sequential mesh reconstruction · SURREAL, Human3.6M, DFAUST | Reconstruction error (mm): 25.7 -> 18.2; 26.5 -> 21.4; 25.9 -> 19.7 | Improved |
| Add temporal mesh attention convolution (TMAC) (ablation) | Proposed sequential mesh reconstruction · SURREAL, Human3.6M, DFAUST | Reconstruction error (mm): 21.2 -> 18.2; 22.9 -> 21.4; 22.2 -> 19.7 | Improved |
| Regress SMPL parameters rather than mesh vertices through the proposed mesh-convolution framework (ablation) | Point-cloud body reconstruction · SURREAL, Human3.6M, DFAUST | Reconstruction error (mm): 18.2 -> 65.8; 21.4 -> 68.4; 19.7 -> 65.3 | Hurt |
| Increase the sampled input point count from 2,500 to 5,000 (ablation) | Proposed sequential mesh reconstruction · SURREAL, Human3.6M, DFAUST | Reconstruction error (mm): 18.2 -> 18.0; 21.4 -> 20.8; 19.7 -> 19.3 | Improved |
| Increase the sampled input point count from 2,500 to 7,500 (ablation) | Proposed sequential mesh reconstruction · SURREAL, Human3.6M, DFAUST | Reconstruction error (mm): 18.2 -> 17.9; 21.4 -> 20.7; 19.7 -> 19.1 | Improved |
| Fine-tune with weak supervision using point correspondences and surface regularizers | Proposed model on real clothed-body point clouds · Real detailed data | 3D model fit to input point clouds | Improved |
Closest papers by meaning
- Human as Points: Explicit Point-Based 3D Human Reconstruction From Single-View RGB Images (TPAMI 2025 · similarity 0.90)
- PointHPS: Cascaded 3D Human Pose and Shape Estimation from Point Clouds (IJCV 2026 · similarity 0.89)
- Convolutional Mesh Regression for Single-Image Human Shape Reconstruction (CVPR 2019 · similarity 0.89)
- Co-Evolution of Pose and Mesh for 3D Human Body Estimation from Video (ICCV 2023 · similarity 0.89)
- Exploiting temporal information for 3D human pose estimation (ECCV 2018 · similarity 0.89)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 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.
