Research map / Video / Video segmentation and tracking
Understanding Human Hands in Contact at Internet Scale
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 |
|---|---|---|---|
| Train the hand detector on 100DOH rather than rely on full-body pose estimation | Hand detection · 100DOH test set | hand detection precision: 43.0% AP; 49.5% recall; 99.7% precision | Helped |
| Use 100DOH training data for hand detection | Faster R-CNN hand detector · Cross-dataset hand detection | relative mAP: 92.9% | Helped |
| Use a hand-specific median box instead of a learned detector | Hand detection on 100DOH · 100DOH | hand detection AP: 0.08% and 0.11% | Hurt |
| Increase hand size from 10% to 20% | Full hand-state system · 100DOH hand-scale evaluation | hand AP: 78.2% to 90.3% | Helped |
| Train mesh-quality assessment with self-consistency-derived positive and negative labels | Hand mesh reconstruction assessment · Mesh quality judgments | AUROC: 90% vs 60% | Helped |
| Replace the self-consistency MLP with Gaussian Naive Bayes while keeping the same training labels | Hand mesh reconstruction assessment · Mesh quality judgments | AUROC: 90% vs 89% | No clear effect |
| Use the detected hand side rather than the opposite side for mesh reconstruction | Hand mesh reconstruction · Hand-side reconstruction judgment | judged correct: 29.1% -> 57.8% | Helped |
| Train the full hand-state system with 90K rather than 15K training examples (ablation) | Full hand-state system · 100DOH | AP for all hand outputs correct: 7% AP | Helped |
| Increase training data from 15K to 45K examples (ablation) | Full hand-state system · 100DOH | AP | Helped |
| Increase training data from 45K to 90K examples (ablation) | Full hand-state system · 100DOH | AP | Helped |
| Predict object-to-hand contact meshes from uncontacted object images | Object-to-grasp system · Object-to-grasp human preference, conclusive judgments | preference for predicted grasp: 72% | Helped |
Closest papers by meaning
- Learning Dense Hand Contact Estimation from Imbalanced Data (NeurIPS 2025 · similarity 0.90)
- How Do I Do That? Synthesizing 3D Hand Motion and Contacts for Everyday Interactions (CVPR 2025 · similarity 0.90)
- HOLD: Category-agnostic 3D Reconstruction of Interacting Hands and Objects from Video (CVPR 2024 · similarity 0.88)
- HO-Cap: A Capture System and Dataset for 3D Reconstruction and Pose Tracking of Hand-Object Interaction (NeurIPS 2025 · similarity 0.88)
- GEARS: Local Geometry-aware Hand-object Interaction Synthesis (CVPR 2024 · similarity 0.88)
- BIGS: Bimanual Category-agnostic Interaction Reconstruction from Monocular Videos via 3D Gaussian Splatting (CVPR 2025 · similarity 0.88)
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.