Research map / 3D vision / Human pose estimation
3DGazeNet: Generalizing Gaze Estimation with Weak Supervision from Synthetic Views
Evangelos Ververas, Polydefkis Gkagkos, Jiankang Deng, Michail C Doukas, Jia Guo, Stefanos Zafeiriou
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 multi-view consistency loss to pseudo-label training (ablation) | 3DGazeNet trained on ITWG · Gaze360 test (cross-dataset) | gaze error (degrees): 23.1 -> 18.1 | Improved |
| add pseudo-label supervision to multi-view loss (ablation) | 3DGazeNet trained on ITWG-MV · Gaze360 test (cross-dataset) | gaze error (degrees): 47.4 -> 18.1 | Improved |
| add multi-view loss on top of GT and pseudo-labels (GC) (ablation) | 3DGazeNet trained on GC+ITWG · Gaze360 test (cross-dataset) | gaze error (degrees): 21.4 -> 17.6 | Improved |
| predict dense eye mesh together with gaze vector instead of vector alone (ablation) | 3DGazeNet within-dataset · Gaze360 within-dataset | gaze error (degrees): 9.8 -> 9.6 | Improved |
| use dense eye mesh instead of 8 iris landmarks in the multi-view generalization setting (ablation) | 3DGazeNet trained on ITWG-MV · Gaze360 test (cross-dataset) | gaze error (degrees): 18.8 -> 18.1 | Improved |
| add mesh target to vector target for generalization (ablation) | 3DGazeNet trained on ITWG-MV · MPII test (cross-dataset) | gaze error (degrees): 8.5 -> 7.6 | Improved |
| add ITWG-MV synthetic-view data to ground-truth training | 3DGazeNet trained on EXG · Gaze360 test (cross-dataset) | gaze error (degrees): 22.1 -> 15.4 | Improved |
| add AVA multi-view supervision to Gaze360 training | 3DGazeNet within-dataset · Gaze360 within-dataset | gaze error (degrees): 9.6 -> 9.7 | No clear effect |
| add large-scale ITWG-MV to Gaze360 training | 3DGazeNet within-dataset · Gaze360 within-dataset | gaze error (degrees): 9.6 -> 9.3 | Improved |
| widen head pose range of in-the-wild training data (yaw up to 90 degrees) (ablation) | 3DGazeNet trained with MPII + ITWG-MV · Gaze360 test across head poses | gaze error (degrees) | Improved |
| add ITWG-MV to training for other baselines without noise regularization | RAT/RUDA, CDG/CRGA, PureGaze · MPII and GC | gaze error (degrees) | No clear effect |
Closest papers by meaning
- Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels (CVPR 2025 · similarity 0.92)
- From Feature to Gaze: A Generalizable Replacement of Linear Layer for Gaze Estimation (CVPR 2024 · similarity 0.91)
- GazeManifold: Learning Canonical Gaze Embeddings for Interpretable and Mapping-Efficient Gaze Estimation (TPAMI 2026 · similarity 0.91)
- GA3CE: Unconstrained 3D Gaze Estimation with Gaze-Aware 3D Context Encoding (CVPR 2025 · similarity 0.90)
- OmniGaze: Reward-inspired Generalizable Gaze Estimation in the Wild (NeurIPS 2025 · similarity 0.90)
- Gaze from Origin: Learning for Generalized Gaze Estimation by Embedding the Gaze Frontalization Process (AAAI 2024 · similarity 0.90)
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.
