Research map / 3D vision / Human pose estimation
Interaction-aware Joint Attention Estimation Using People Attributes
Chihiro Nakatani, Hiroaki Kawashima, Norimichi Ukita
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 interaction-aware PJAT with people attributes to image-based DAVT (ablation) | DAVT · Volleyball (Ex.1) | Dist (pixels): 72.0 -> 56.0 | Improved |
| add image-based DAVT branch to PJAT (ablation) | PJAT · Volleyball (Ex.1) | Dist (pixels): 87.2 -> 56.0 | Improved |
| add gaze direction attribute (ablation) | PJAT · Volleyball (Ex.1) | Dist (pixels): 70.8 -> 56.0 | Improved |
| add person location attribute (ablation) | PJAT · Volleyball (Ex.1) | Dist (pixels): 60.3 -> 56.0 | Improved |
| add predicted action attribute (ablation) | PJAT · Volleyball (Ex.1) | Dist (pixels): 60.2 -> 56.0 fused; branch alpha 82.1 vs 84.8 | No clear effect |
| add ground-truth action attribute (ablation) | PJAT · Volleyball (Ex.2) | Dist (pixels): 39.2 -> 11.4 | Improved |
| predict heatmap pixelwise with positional embedding instead of imagewise (ablation) | PJAT heatmap head · Volleyball (Ex.1) | Dist (alpha branch): 126.7 -> 87.5 | Improved |
| use learned weighted fusion instead of CNN fusion (ablation) | fusion module · Volleyball (Ex.1) | Dist (pixels): 94.6 -> 56.0 | Improved |
| use learned weighted fusion instead of simple average (ablation) | fusion module · Volleyball (Ex.1) | Dist (pixels): 58.1 -> 56.0 | Improved |
| predict J_JA directly instead of averaging per-person heatmaps (ablation) | PJAT · Volleyball (Ex.1) | Dist (alpha branch): 87.8 -> 87.5 | No clear effect |
| use predicted rather than ground-truth gaze at test time (ablation) | PJAT trained on ground truth · Volleyball (Ex.2 error analysis) | Dist (pixels): 11.4 -> 113.1 | Hurt |
| use predicted rather than ground-truth action at test time (ablation) | PJAT trained on ground truth · Volleyball (Ex.2 error analysis) | Dist (pixels): 11.4 -> 19.4 | Hurt |
Closest papers by meaning
- Connecting Gaze, Scene, and Attention: Generalized Attention Estimation via Joint Modeling of Gaze and Scene Saliency (ECCV 2018 · similarity 0.89)
- In the Eye of Beholder: Joint Learning of Gaze and Actions in First Person Video (ECCV 2018 · similarity 0.87)
- Learning from Observer Gaze: Zero-Shot Attention Prediction Oriented by Human-Object Interaction Recognition (CVPR 2024 · similarity 0.87)
- Unifying Top-down and Bottom-up Scanpath Prediction Using Transformers (CVPR 2024 · similarity 0.87)
- Sharingan: A Transformer Architecture for Multi-Person Gaze Following (CVPR 2024 · similarity 0.86)
- Gaze-LLE: Gaze Target Estimation via Large-Scale Learned Encoders (CVPR 2025 · similarity 0.84)
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.
