Research map / 3D vision / Human pose estimation
Efficiently Creating 3D Training Data for Fine Hand Pose Estimation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Select reference frames by appearance-based submodular coverage instead of regular temporal sampling (ablation) | Reference-frame selection · Synthetic sequence | average 3D annotation error: 6.45\,mm -> 5.53\,mm | Helped |
| Select reference frames by appearance-based submodular coverage instead of regular temporal sampling (ablation) | Reference-frame selection · Synthetic sequence | manual interventions: 276 -> 133 | Helped |
| Add visibility constraints to 2D reprojection and bone-length constraints (ablation) | Reference-frame initialization · Synthetic sequence, all joints | average 3D joint error: 19.98 / 13.29\,mm -> 6.20 / 3.41\,mm | Helped |
| Add z-order constraints to 2D reprojection, visibility, and bone-length constraints (ablation) | Reference-frame initialization · Synthetic sequence, all joints | average 3D joint error: 6.20 / 3.41\,mm -> 3.65 / 2.98\,mm | Helped |
| Add Gaussian noise to 2D joint locations (ablation) | Reference-frame initialization · Synthetic sequence, all joints | average 3D joint error: 3.65\,mm -> $4.29\pm 0.63$\,mm | Hurt |
| Add SIFTFlow alignment before frame optimization (ablation) | 3D location propagation · Synthetic sequence | average 3D joint error: 11.50 -> 11.40\,mm | No clear effect |
| Optimize propagated joint locations within each frame (ablation) | 3D location propagation · Synthetic sequence | average 3D joint error: 11.40 -> 5.76\,mm | Helped |
| Add global temporal optimization after frame-level optimization (ablation) | 3D location propagation · Synthetic sequence | average 3D joint error: 5.76 -> 5.53\,mm | Helped |
| Train and test the pose estimator using revised annotations rather than provided annotations | State-of-the-art 3D hand pose estimator · MSRA, 10-fold cross-validation | test set error: $6.41\pm 2.05$\,mm -> $5.58\pm 0.56$\,mm | Helped |
| Train on provided annotations and evaluate against revised annotations | State-of-the-art 3D hand pose estimator · MSRA, 10-fold cross-validation | test set error: $13.23\pm 6.98$\,mm | Hurt |
Closest papers by meaning
- Dense 3D Regression for Hand Pose Estimation (CVPR 2018 · similarity 0.93)
- Depth-Based 3D Hand Pose Estimation: From Current Achievements to Future Goals (CVPR 2018 · similarity 0.92)
- Self-Supervised 3D Hand Pose Estimation Through Training by Fitting (CVPR 2019 · similarity 0.92)
- DeepHand: Robust Hand Pose Estimation by Completing a Matrix Imputed With Deep Features (CVPR 2016 · similarity 0.92)
- SiMHand: Mining Similar Hands for Large-Scale 3D Hand Pose Pre-training (ICLR 2025 · similarity 0.90)
- Analyzing the Synthetic-to-Real Domain Gap in 3D Hand Pose Estimation (CVPR 2025 · similarity 0.89)
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.