Research map / 3D vision / Human pose estimation
Forecasting 3D Scanpaths in Egocentric Video
Fiona Ryan, Ishwarya Ananthabhotla, Yijun Qian, Judy Hoffman, James M. Rehg, Vamsi Krishna Ithapu, Calvin Murdock
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 |
|---|---|---|---|
| add visual context from the canonical frame (ablation) | 3D scanpath transformer · ADT | DTW: 1.450 -> 1.410 | Improved |
| add past video frames and head poses to single-image model (ablation) | 3D scanpath transformer · ADT | DTW: 1.410 -> 1.377 | Improved |
| add head pose to video model (ablation) | 3D scanpath transformer · ADT | Z error (m): 0.918 -> 0.890 | Improved |
| add head pose to single-image model (ablation) | 3D scanpath transformer · ADT | Dur (MultiMatch): 0.671 -> 0.698 | Hurt |
| provide 1 past fixation as context instead of none (ablation) | 3D scanpath transformer · ADT | DTW: 1.720 -> 1.450 | Improved |
| increase context from 1 to 10 fixations (ablation) | 3D scanpath transformer · ADT | DTW: 1.450 -> 1.377 | Improved |
| use bidirectional instead of causal decoder attention (ablation) | 3D scanpath transformer · ADT | DTW: 1.419 -> 1.377 | Improved |
| bidirectional vs partial causal attention (ablation) | 3D scanpath transformer · ADT | TDE: 0.854 vs 0.859 | No clear effect |
| train in 3D instead of 2D then project to 2D (ablation) | 3D scanpath transformer · ADT 2D projection | DTW: 0.115 -> 0.088 | Improved |
| adapt TPP-Gaze to 3D prediction | TPP-Gaze · ADT 2D projection | DTW: 0.121 -> 0.128 | Hurt |
| fine-tune TPP-Gaze on ADT instead of zero-shot | TPP-Gaze · ADT 2D projection | DTW: 0.151 -> 0.121 | Improved |
| use linear extrapolation instead of last observed point | heuristic baseline · ADT | DTW: 1.533 -> 4.642 | Hurt |
Closest papers by meaning
- Gaze Beyond the Frame: Forecasting Egocentric 3D Visual Span (NeurIPS 2025 · similarity 0.93)
- Ego-Human Motion Prediction with 3D-Aware LLM (ECCV 2026 · similarity 0.89)
- EgoGen: An Egocentric Synthetic Data Generator (CVPR 2024 · similarity 0.87)
- EgoSim: An Egocentric Multi-view Simulator and Real Dataset for Body-worn Cameras during Motion and Activity (NeurIPS 2024 · similarity 0.87)
- Instance Tracking in 3D Scenes from Egocentric Videos (CVPR 2024 · similarity 0.86)
- GazeFlow: From Human Gaze Behavior to Generative Egocentric Gaze Prediction (NeurIPS 2026 · similarity 0.86)
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.
