Research map / 3D vision / 3D generation
Seeing the Unseen: Physics-as-Representation for Generalizable Gaze Perception
Yunfeng Xiao, Xiaowei Bai, Hao Su, Hao He, Liang Xie, Erwei Yin
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 MGR branch (3D edge-supervised reconstruction) (ablation) | SG-Gaze · TEyeD DT1->DT2 | 3D gaze error: 3.11 -> 1.88 | Improved |
| add AGE branch (analytical spherical gaze estimation) (ablation) | SG-Gaze · TEyeD DT1->DT2 | 3D gaze error: 2.57 -> 1.88 | Improved |
| add View-Consistent Regularization (ablation) | SG-Gaze · TEyeD DT1->DT2 | 3D gaze error: 2.07 -> 1.88 | Improved |
| add MGR branch (ablation) | SG-Gaze · TEyeD within DT1 | 3D gaze error: 2.86 -> 0.94 | Improved |
| add AGE branch (ablation) | SG-Gaze · TEyeD within DT1 | 3D gaze error: 1.83 -> 0.94 | Improved |
| add VCR (ablation) | SG-Gaze · TEyeD within DT1 | 3D gaze error: 1.53 -> 0.94 | Improved |
| add SG-Gaze to ResNet-18 | ResNet-18 · TEyeD DT1->DT2 | 3D gaze error: 3.20 -> 2.07 | Improved |
| add SG-Gaze to ResNet-50 | ResNet-50 · TEyeD DT1->LPW | 3D gaze error: 5.47 -> 3.75 | Improved |
| add SG-Gaze to DINOv2 ViT-B/14 | DINOv2 (ViT-B/14) · TEyeD DT1->DT2 | 3D gaze error: 2.09 -> 1.76 | Improved |
| add SG-Gaze to VGG-16 | VGG-16 · TEyeD within DT1 | 3D gaze error: 2.12 -> 2.78 | Hurt |
| add SG-Gaze to ResNet-50 | ResNet-50 · TEyeD within DT1 | 3D gaze error: 1.32 -> 1.22 | Improved |
| add semantic/centre losses (Sem.+Gaze+Cent.) to gaze-only training (ablation) | SG-Gaze · TEyeD DT1->DT2 | 3D gaze error: 1.88 -> 2.41 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- De2Gaze — De^2Gaze: Deformable and Decoupled Representation Learning for 3D Gaze Estimation (CVPR 2025) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- From Feature to Gaze: A Generalizable Replacement of Linear Layer for Gaze Estimation (CVPR 2024 · similarity 0.85)
- De^2Gaze: Deformable and Decoupled Representation Learning for 3D Gaze Estimation (CVPR 2025 · similarity 0.85)
- GEOPHYS: The Geometry of Physical Plausibility (NeurIPS 2026 · similarity 0.85)
- OmniGaze: Reward-inspired Generalizable Gaze Estimation in the Wild (NeurIPS 2025 · similarity 0.85)
- Gaze Beyond the Frame: Forecasting Egocentric 3D Visual Span (NeurIPS 2025 · similarity 0.85)
- GA3CE: Unconstrained 3D Gaze Estimation with Gaze-Aware 3D Context Encoding (CVPR 2025 · similarity 0.85)
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.
