Research map / Point clouds / Point cloud classification
Sparkle: A Robust and Versatile Representation for Point Cloud-based Human Motion Capture
Yiming Ren, Yujing Sun, Aoru Xue, Kwok-Yan Lam, Yuexin Ma
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 Point-aligned Skeleton Tracker (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 149.2/159.6 -> 105.1/113.9 | Improved |
| add Skeleton-guided Anchor Estimator (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 116.0/125.3 -> 105.1/113.9 | Improved |
| add Sparkle-based SMPL Solver (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 112.6/122.4 -> 105.1/113.9 | Improved |
| use residual offset instead of direct prediction in PST (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 107.2/116.7 -> 105.1/113.9 | Improved |
| add linear initialization in SAE (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 108.3/118.4 -> 105.1/113.9 | Improved |
| use cross-attention refinement rather than MLP refinement in SAE (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 106.9/115.6 -> 105.1/113.9 | Improved |
| use 32 PCA anchors instead of 16 or 64/96 (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 112.7/121.0 (16), 107.4/115.3 (64) -> 105.1/113.9 (32) | Improved |
| use PCA anchors instead of random anchors (ablation) | SparkleMotion · FreeMotion | global J/V error (mm): 117.2/126.3 -> 105.1/113.9 | Improved |
| use manual CMU marker anchors instead of PCA anchors (ablation) | SparkleMotion · HuMMan-Point | global J/V error (mm): 87.5/97.6 -> 85.3/96.4 (60 anchors) | No clear effect |
| replace residual offset with 2-stage iterative refinement in PST (ablation) | SparkleMotion · FreeMotion | local J/V error (mm): 59.0/73.2 -> 58.5/72.8 | No clear effect |
| increase point occlusion ratio from 0% to 90% | SparkleMotion · FreeMotion | global J/V error (mm): 105.1/113.9 -> 163.3/179.0 | Hurt |
| use Sparkle representation vs LiveHPS++ with no target-domain data | SparkleMotion · FreeMotion (cross-dataset) | local joint error (mm): 84.0 -> 65.2 | Improved |
Closest papers by meaning
- DANCE: Density-agnostic and Class-aware Network for Point Cloud Completion (AAAI 2026 · similarity 0.85)
- MotionBERT: A Unified Perspective on Learning Human Motion Representations (ICCV 2023 · similarity 0.84)
- Learnable Skeleton-Aware 3D Point Cloud Sampling (CVPR 2023 · similarity 0.84)
- MultiPhys: Multi-Person Physics-aware 3D Motion Estimation (CVPR 2024 · similarity 0.83)
- 3DInAction: Understanding Human Actions in 3D Point Clouds (CVPR 2024 · similarity 0.83)
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 2025 · similarity 0.83)
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.
