Research map / 3D vision / Face and shape
What we need is explicit controllability: Training 3D gaze estimator using only facial images
Tingwei Li, Jun Bao, Zhenzhong Kuang, Buyu Liu
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 random symmetric rotation (RSR) of eyeballs during training (ablation) | Gaze synthesis pipeline · MPII | 3D angular error (deg): 7.84 -> 7.39 | Improved |
| add random symmetric rotation (RSR) (ablation) | Gaze synthesis pipeline · MPII | FID: 18.39 -> 19.95 | Hurt |
| add eye geometry loss Lg (ablation) | Gaze synthesis pipeline · MPII | 3D angular error (deg): 7.97 -> 7.39 | Improved |
| add virtual screen gaze target (ablation) | Gaze synthesis pipeline · MPII | 3D angular error (deg): 8.03 -> 7.39 | Improved |
| add eye appearance loss Le (ablation) | Gaze synthesis pipeline · MPII | 3D angular error (deg): 8.49 -> 7.39 | Improved |
| add eye appearance loss Le (ablation) | Gaze synthesis pipeline · MPII | FID: 29.29 -> 19.95 | Improved |
| add learnable lighting model Omega instead of spherical harmonics color (ablation) | Gaze synthesis pipeline · MPII | FID: 27.65 -> 19.95 | Improved |
| generate full facial images rather than eye patches only (ablation) | Gaze synthesis pipeline · MPII | 3D angular error (deg): 7.48 -> 7.39 | Improved |
| add all proposed components to the GaussianAvatars baseline (ablation) | Gaussian-avatar baseline · MPII | 3D angular error (deg): 8.25 -> 7.39 | Improved |
| increase generated images per subject beyond about 150 | Gaze synthesis pipeline · MPII | gaze error | No clear effect |
| proposed method vs best unsupervised baseline on EVE | Gaze synthesis pipeline · EVE test | 3D angular error (deg): 11.10 -> 9.00 | Improved |
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.
- GaussianAvatars — GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians (CVPR 2024) · better on 2 of 2 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
- GazeGaussian: High-Fidelity Gaze Redirection with 3D Gaussian Splatting (ICCV 2025 · similarity 0.90)
- OmniGaze: Reward-inspired Generalizable Gaze Estimation in the Wild (NeurIPS 2025 · similarity 0.89)
- Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels (CVPR 2025 · similarity 0.89)
- GA3CE: Unconstrained 3D Gaze Estimation with Gaze-Aware 3D Context Encoding (CVPR 2025 · similarity 0.88)
- 3D Prior Is All You Need: Cross-Task Few-shot 2D Gaze Estimation (CVPR 2025 · similarity 0.88)
- De^2Gaze: Deformable and Decoupled Representation Learning for 3D Gaze Estimation (CVPR 2025 · similarity 0.87)
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.
