Research map / 3D vision / Human pose estimation
Dense 3D Regression for Hand Pose Estimation
Read the paperCheck 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 |
|---|---|---|---|
| use dense per-pixel joint heat maps and 3D offset directions with consensus aggregation instead of holistic pose regression | hand pose estimation · NYU | average 3D error (mm): 11.8mm -> 10.2mm | Helped |
| use dense per-pixel joint heat maps and 3D offset directions with consensus aggregation instead of holistic pose regression | hand pose estimation · MSRA | average 3D error (mm): 8.6mm -> 7.2mm | Helped |
| use dense per-pixel joint heat maps and 3D offset directions with consensus aggregation instead of holistic pose regression | hand pose estimation · ICVL | average 3D error (mm): 6.8mm -> 7.3mm | Hurt |
| add 2D joint detection to coordinate regression | holistic 3D regression · NYU | average joint error (mm): 15.04mm -> 14.88mm | Helped |
| use 2D detections to define x,y and regress only depth instead of regressing full 3D coordinates | holistic 3D regression · NYU | average joint error (mm): 14.88mm -> 13.91mm | Helped |
| replace holistic coordinate regression with dense offset vector-field regression | hand pose estimation · NYU | average joint error (mm): 15.04mm -> 12.56mm | Helped |
| add a second network stack (ablation) | Dense 3D Regression · NYU | average 3D error (mm): 11.20 -> 10.21 | Helped |
| use 5 mean-shift candidate points instead of 1 (ablation) | Dense 3D Regression · NYU | average 3D error (mm): 10.6 -> 10.21 | Helped |
| use 10 or 30 mean-shift candidate points instead of 5 (ablation) | Dense 3D Regression · NYU | average 3D error (mm): 5: 10.21; 10: 10.21; 30: 10.96 | Mixed |
| set mean-shift kernel width to 10mm instead of 40mm (ablation) | Dense 3D Regression · NYU | average 3D error (mm): 10.21 -> 10.35 | Hurt |
| set mean-shift kernel width to 80mm or 100mm instead of 40mm (ablation) | Dense 3D Regression · NYU | average 3D error (mm): 40mm: 10.21; 80mm: 10.21; 100mm: 10.21 | No clear effect |
| remove weighted mean-shift aggregation (ablation) | Dense 3D Regression · NYU | average 3D error (mm): 10.21 -> 10.26 | Hurt |
Closest papers by meaning
- Robust 3D Hand Pose Estimation in Single Depth Images: From Single-View CNN to Multi-View CNNs (CVPR 2016 · similarity 0.94)
- Neural Voting Field for Camera-Space 3D Hand Pose Estimation (CVPR 2023 · similarity 0.93)
- Efficiently Creating 3D Training Data for Fine Hand Pose Estimation (CVPR 2016 · similarity 0.93)
- Depth-Based 3D Hand Pose Estimation: From Current Achievements to Future Goals (CVPR 2018 · similarity 0.93)
- Self-Supervised 3D Hand Pose Estimation Through Training by Fitting (CVPR 2019 · similarity 0.92)
- Keypoint Fusion for RGB-D Based 3D Hand Pose Estimation (AAAI 2024 · similarity 0.90)
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.