Research map / 3D vision / Human pose estimation
Real-Time Hand Tracking Under Occlusion From an Egocentric RGB-D Sensor
Franziska Mueller, Dushyant Mehta, Oleksandr Sotnychenko, Srinath Sridhar, Dan Casas, Christian Theobalt
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 |
|---|---|---|---|
| regress heatmaps for all joints (auxiliary task) in HALNet instead of the root only (ablation) | HALNet · SynthHands test set | 2D root pixel error (px): 2.35 px -> 2.2 px (6.4%) | Improved |
| use reduced 10-block HALNet instead of ResNet50 for hand localization (ablation) | HALNet · SynthHands test set | 2D root pixel error (px): 2.1 px -> 2.2 px | No clear effect |
| use reduced HALNet instead of ResNet50 (speed) (ablation) | HALNet · SynthHands, GPU forward pass | forward time (ms): 18 ms -> 10 ms | Improved |
| use reduced JORNet instead of ResNet50 (speed) (ablation) | JORNet · SynthHands, GPU forward pass | forward time (ms): 11 ms -> 6 ms | Improved |
| train JORNet on all joints rather than only fingertips and wrist (JORNet light) (ablation) | JORNet · SynthHands test set | 3D wrist and fingertip error | Improved |
| use ground truth crops instead of hand-localization crops for JORNet (ablation) | JORNet · SynthHands test set | 3D joint error | Improved |
| use hand-localization crops instead of ground truth crops for JORNet (ablation) | JORNet · SynthHands test set | 2D joint error | Improved |
| split into two CNNs (localization then regression) instead of one CNN on the full frame (ablation) | two-step RGB-D CNN · EgoDexter | 3D fingertip error | Improved |
| add RGB input to depth for the two-step CNNs (ablation) | two-step CNN · EgoDexter | 3D fingertip error | Improved |
| add kinematic tracking with both 2D and 3D terms instead of only 3D term (ablation) | kinematic pose tracking · EgoDexter | average fingertip error: 32.6 mm | Improved |
| use only the 2D fitting energy in tracking (ablation) | kinematic pose tracking · EgoDexter | average fingertip error | Hurt |
Closest papers by meaning
- Real-Time Joint Tracking of a Hand Manipulating an Object from RGB-D Input (ECCV 2016 · similarity 0.91)
- H+O: Unified Egocentric Recognition of 3D Hand-Object Poses and Interactions (CVPR 2019 · similarity 0.91)
- ComPose: When to Trust Hands for Object Pose Tracking (NeurIPS 2026 · similarity 0.89)
- HOT3D: Hand and Object Tracking in 3D from Egocentric Multi-View Videos (CVPR 2025 · similarity 0.89)
- EgoPressure: A Dataset for Hand Pressure and Pose Estimation in Egocentric Vision (CVPR 2025 · similarity 0.88)
- StableHand: Quality-Aware Flow Matching for World-Space Dual-Hand Motion Estimation from Egocentric Video (NeurIPS 2026 · 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.
