Research map / 3D vision / Face and shape
Tracking-by-Detection of 3D Human Shapes: From Surfaces to Volumes
Chun-Hao P. Huang, Benjamin Allain, Edmond Boyer, Jean-Sébastien Franco, Federico Tombari, Nassir Navab, Slobodan Ilić
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 |
|---|---|---|---|
| use random-forest discriminative associations instead of ICP (surICP) at low frame rate | surface tracking · Jumping (every third frame) | silhouette overlap error: 16759.29 -> 9148.94 | Improved |
| use discriminative associations instead of surICP at low frame rate | surface tracking · Crane (every third frame) | silhouette overlap error: 8295.58 -> 7746.40 | Improved |
| use discriminative associations instead of surICP at low frame rate | surface tracking · Bouncing (every third frame) | silhouette overlap error: 9400.76 -> 6847.72 | Improved |
| use discriminative associations instead of surICP at low frame rate | surface tracking · Handstand (every third frame) | silhouette overlap error: 11690.61 -> 9279.57 | Improved |
| use discriminative associations instead of surICP at full frame rate | surface tracking · sequences without outliers | silhouette overlap error | No clear effect |
| use CVT-based fully volumetric tracking instead of voxel-based pipeline [11] | tracking-by-detection · Thomas/Seq2 | silhouette pixel error mean: 28731 -> 9991 | Improved |
| use CVT-based tracking instead of voxel-based [11] | tracking-by-detection · Ballet/Seq2 | silhouette pixel error mean: 5427 -> 2620 | Improved |
| use CVT-based tracking instead of voxel-based [11] | tracking-by-detection · Goalkeeper/UpJump | silhouette pixel error mean: 19838 -> 15221 | Improved |
| use CVT-based tracking instead of voxel-based [11] | tracking-by-detection · Ballet/Seq2 markers | mean marker error (mm): 124.02 -> 26.37 | Improved |
| use proposed instead of ICP-based Allain et al. [51] | tracking-by-detection · Goalkeeper/UpJump | silhouette pixel error mean: 14773 -> 15221 | No clear effect |
| use proposed instead of Allain et al. [51] | tracking-by-detection · Ballet/Seq2 markers | mean marker error (mm): 27.82 -> 26.37 | Improved |
| train forest across all three subjects (cross-template) instead of Goalkeeper only (ablation) | CVT forest · Goalkeeper/UpJump correspondences | correspondence accuracy | No clear effect |
Closest papers by meaning
- Volumetric 3D Tracking by Detection (CVPR 2016 · similarity 0.89)
- Capturing Humans in Motion: Temporal-Attentive 3D Human Pose and Shape Estimation From Monocular Video (CVPR 2022 · similarity 0.89)
- NPMs: Neural Parametric Models for 3D Deformable Shapes (ICCV 2021 · similarity 0.89)
- Recovering 3D Human Mesh From Monocular Images: A Survey (TPAMI 2023 · similarity 0.89)
- Convolutional Mesh Regression for Single-Image Human Shape Reconstruction (CVPR 2019 · similarity 0.88)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.87)
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.
