Research map / Label-efficient and robust learning / Domain adaptation and generalization
De-confounded Gaze Estimation
Ziyang Liang, Yiwei Bao, Feng Lu
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 |
|---|---|---|---|
| apply FSCI causal intervention framework to ResNet18 | ResNet18 gaze estimator · DE->DM | angular error (deg): 9.07 -> 5.79 | Improved |
| add Causal Intervention Module on top of Feature Separation Module (ablation) | FSCI-ResNet18 · DE->DM | angular error (deg): 7.67 -> 5.79 | Improved |
| use cross-attention Feature Separation Module instead of three separate ResNet18 (ablation) | FSCI-ResNet18 · DE->DM | angular error (deg): 8.72 -> 5.79 | Improved |
| apply FSCI to a ResNet50 backbone | ResNet50 · DG->DD | angular error (deg): 11.83 -> 10.20 | Improved |
| add CIM to FSCI-ResNet50 (versus w/o CIM) (ablation) | FSCI-ResNet50 · DE->DD | angular error (deg): DE->DD 7.23 -> 6.68; DE->DM 6.17 -> 5.47 | Mixed |
| use FSM with separate ResNet50x3 instead of cross-attention separation (ablation) | FSCI-ResNet50 · DG->DD | angular error (deg): 10.20 -> 12.99 | Hurt |
| use Dynamic Confounder Bank instead of confounder dictionary (ablation) | FSCI · DE->DM | angular error (deg): 5.77 -> 5.79 (best dictionary vs FSCI) | No clear effect |
| use Dynamic Confounder Bank instead of confounder dictionary (ablation) | FSCI · DE->DD | angular error (deg): 8.18 -> 6.96 | Improved |
| apply Gaussian noise (std 0.2) to input; compare FSCI to baseline robustness | FSCI vs ResNet18 · DE->DM | error increase (deg): 4.13 -> 1.29 | Improved |
| apply causal intervention to reduce head-pose-to-gaze spurious correlation | FSCI · DE->DM, complexion groups | error difference between complexions: 1.26 vs 0.72 deg difference | 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.
- PureGaze — PureGaze: Purifying Gaze Feature for Generalizable Gaze Estimation (AAAI 2022) · better on 1 of 1 shared result
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
- PureGaze: Purifying Gaze Feature for Generalizable Gaze Estimation (AAAI 2022 · similarity 0.90)
- Gaze from Origin: Learning for Generalized Gaze Estimation by Embedding the Gaze Frontalization Process (AAAI 2024 · similarity 0.90)
- From Feature to Gaze: A Generalizable Replacement of Linear Layer for Gaze Estimation (CVPR 2024 · similarity 0.88)
- Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels (CVPR 2025 · similarity 0.87)
- Gaze Label Alignment: Alleviating Domain Shift for Gaze Estimation (AAAI 2025 · similarity 0.87)
- CLIP-Gaze: Towards General Gaze Estimation via Visual-Linguistic Model (AAAI 2024 · similarity 0.86)
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.
