Research map / Video / Video segmentation and tracking
Joint Hand Motion and Interaction Hotspots Prediction From Egocentric Videos
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 |
|---|---|---|---|
| Use stochastic C-VAE rather than deterministic MLP trajectory head (ablation) | OCT · EK100 validation | ADE: 0.21 -> 0.12 | Helped |
| Use stochastic C-VAE rather than deterministic MLP hotspots head (ablation) | OCT · EK100 validation | NCC: 0.43 -> 0.72 | Helped |
| Use C-VAE latent-space modeling rather than Bivariate output-space modeling (ablation) | OCT trajectory prediction · EK100 validation | ADE: 0.19 -> 0.12 | Helped |
| Use C-VAE rather than MDN for contact points (ablation) | OCT hotspots prediction · EK100 validation | NCC: 0.53 -> 0.72 | Helped |
| Condition object contact prediction on future hand trajectory (ablation) | OCT · EK100 validation | NCC: 0.54 -> 0.72 | Helped |
| Train on the larger automatically annotated EK100 set instead of EK55 (ablation) | OCT · EK100 validation | NCC: 0.60 -> 0.72 | Helped |
| Include global environment features (ablation) | OCT trajectory prediction · EK100 validation | ADE: 0.15 -> 0.12 | Helped |
| Include hand features (ablation) | OCT trajectory prediction · EK100 validation | FDE: 0.16 -> 0.11 | Helped |
| Fine-tune the backbone end-to-end rather than freeze it (ablation) | OCT · EK100 validation | ADE: 0.12 -> 0.11 | No clear effect |
| Use OCT rather than I3D 3D CNN | Future interaction prediction · EK100 validation | ADE: 0.19 -> 0.12 | Helped |
| Pre-train the encoder on trajectory and hotspots prediction before action anticipation | OCT action anticipation · EK55 validation | Action top-5 accuracy: 18.9 -> 24.4 | Helped |
| Use joint attention across tokens instead of divided temporal and spatial attention | OCT hotspots prediction · EK100 validation | NSS: 0.50 -> 0.72 | Helped |
Closest papers by meaning
- How Do I Do That? Synthesizing 3D Hand Motion and Contacts for Everyday Interactions (CVPR 2025 · similarity 0.90)
- Modeling Fine-Grained Hand-Object Dynamics for Egocentric Video Representation Learning (ICLR 2025 · similarity 0.90)
- MEgoHand: Multimodal Egocentric Hand-Object Interaction Motion Generation (NeurIPS 2025 · similarity 0.89)
- Hand-Centric Motion Refinement for 3D Hand-Object Interaction via Hierarchical Spatial-Temporal Modeling (AAAI 2024 · similarity 0.89)
- HOT3D: Hand and Object Tracking in 3D from Egocentric Multi-View Videos (CVPR 2025 · similarity 0.89)
- GEARS: Local Geometry-aware Hand-object Interaction Synthesis (CVPR 2024 · 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.