Research map / Point clouds / Registration and matching
SOMA: Solving Optical Marker-Based MoCap Automatically
Nima Ghorbani, Michael J. Black
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 AMASS noise model to synthetic data (ablation) | SOMA · HDM05 | accuracy: 94.73 -> 95.50 | Improved |
| add CAESAR bodies for body shape variation (ablation) | SOMA · HDM05 | accuracy: 95.21 -> 95.50 | Improved |
| use Sinkhorn optimal transport instead of Log-Softmax (ablation) | SOMA · HDM05 | accuracy: 91.51 -> 95.50 | Improved |
| add random marker placement noise (ablation) | SOMA · HDM05 | accuracy: 89.41 -> 95.50 | Improved |
| add transformer self-attention layers (ablation) | SOMA · HDM05 | accuracy: 11.36 -> 95.50 | Improved |
| use tracklet labeling instead of per-frame labeling | SOMA · BMLmovi | accuracy: 95.90 -> 97.33 | Improved |
| use tracklet labeling instead of per-frame labeling | SOMA · KIT | accuracy: 94.97 -> 95.46 | Improved |
| train with full noise (occlusion and ghost) instead of base noise, tested on noisy data | SOMA · HDM05 (B+C+G test) | accuracy: 73.95 -> 95.50 | Improved |
| train with full noise instead of base noise, tested on clean data | SOMA · HDM05 (B test) | accuracy: 97.93 -> 95.74 | Hurt |
| train on varied AMASS bodies and motions (SOMA*) instead of dataset-specific synthetic markers | SOMA · BMLrub (5 occlusions + ghost) | accuracy: 95.13 -> 97.56 | Improved |
| train one model per reduced layout instead of a base model on the full layout | SOMA · HDM05 (3 markers removed) | accuracy: 95.35 -> 91.78 | Hurt |
| train on real instead of synthetic markers (same motions) | SOMA · BMLrub (0 occlusions) | accuracy: 99.16 -> 99.08 | No clear effect |
Closest papers by meaning
- AMASS: Archive of Motion Capture As Surface Shapes (ICCV 2019 · similarity 0.88)
- Self-supervised Learning of Motion Capture (NeurIPS 2017 · similarity 0.86)
- 3D Human Mesh Estimation From Virtual Markers (CVPR 2023 · similarity 0.86)
- MoCapAnything: Unified 3D Motion Capture for Arbitrary Skeletons from Monocular Videos (CVPR 2026 · similarity 0.86)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.84)
- MotionPRO: Exploring the Role of Pressure in Human MoCap and Beyond (CVPR 2025 · similarity 0.84)
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.
