Research map / 3D vision / Face and shape
Real Time Eye Gaze Tracking With 3D Deformable Eye-Face Model
Kang Wang, Qiang Ji
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 |
|---|---|---|---|
| use DEFM (eye-face model without head-eye offset) instead of DFM (ablation) | gaze estimation with individualized model · Real data, 10 subjects | gaze error (degrees): 5.7 -> 3.5 | Improved |
| use DEFM instead of DFM with the average model (ablation) | gaze estimation · Real data, 10 subjects (average model) | gaze error (degrees): 6.5 -> 4.5 | Improved |
| use personal calibration to individualize the model instead of the average model (ablation) | DEFM · Real data, 10 subjects | gaze error (degrees): 4.5 -> 3.5 | Improved |
| add eye corners to the landmark set (config 2 vs config 1) (ablation) | DEFM · Real data, 10 subjects | gaze error (degrees): 4.1 -> 3.5 | Improved |
| add eyebrow landmarks to the set (config 2 vs config 5) (ablation) | DEFM · Real data, 10 subjects | gaze error (degrees): 4.8 -> 3.5 | Improved |
| use manual pupil annotation instead of pupil detector from [30] (ablation) | DEFM · Real data, 10 subjects | gaze error (degrees): 3.5 -> 2.5 | Improved |
| use starburst pupil detector [29] instead of the fast detector (ablation) | DEFM · Real data, 10 subjects | gaze error (degrees) and time (ms): 3.5 -> 3.1 degrees; 2.3 -> 30.0 ms | Mixed |
| use DEFM instead of DFM under head pose change (ablation) | simulated data, different head pose · Simulated data, 1 pixel noise | testing error e3 (degrees): 12.4 vs 13.5 | Improved |
| use DEFM instead of DFM under 2D landmark noise (ablation) | simulated data · Simulated data, landmark noise sweep | gaze error (degrees): margin of 5 degree | Improved |
| use DEFM instead of DFM for calibration cost (ablation) | calibration · Simulated data | calibration time: 40x more time for DFM | Improved |
| propose DEFM over prior landmark-based methods | gaze estimation · ColumbiaGaze | gaze error (degrees): 8.9 -> 7.1 | Improved |
| propose DEFM over prior methods on head movement | gaze estimation · EyeDiap (HD camera, head movement) | gaze error (degrees): 22.2 -> 16.5 | Improved |
Closest papers by meaning
- What we need is explicit controllability: Training 3D gaze estimator using only facial images (ICCV 2025 · similarity 0.88)
- Dynamic 3D Gaze From Afar: Deep Gaze Estimation From Temporal Eye-Head-Body Coordination (CVPR 2022 · similarity 0.87)
- Monocular Free-Head 3D Gaze Tracking With Deep Learning and Geometry Constraints (ICCV 2017 · similarity 0.86)
- De^2Gaze: Deformable and Decoupled Representation Learning for 3D Gaze Estimation (CVPR 2025 · similarity 0.86)
- 3D Prior Is All You Need: Cross-Task Few-shot 2D Gaze Estimation (CVPR 2025 · similarity 0.85)
- Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels (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.
