Research map / 3D vision / Human pose estimation
Gaze360: Physically Unconstrained Gaze Estimation in the Wild
Petr Kellnhofer, Adria Recasens, Simon Stent, Wojciech Matusik, Antonio Torralba
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 temporal input (LSTM over 7 frames) instead of single frame (ablation) | Pinball loss model · Gaze360 test (all 360 deg) | mean angular error (deg): 15.6 -> 13.5 | Improved |
| use bidirectional LSTM instead of TRN for temporal aggregation (ablation) | Pinball loss model · Gaze360 test (all 360 deg) | mean angular error (deg): 14.1 -> 13.5 | Improved |
| add pinball loss quantile regression instead of MSE (ablation) | LSTM model · Gaze360 test (all 360 deg) | mean angular error (deg): 14.1 -> 13.5 | Improved |
| add pinball loss instead of MSE (ablation) | Static model · Gaze360 test (all 360 deg) | mean angular error (deg): 15.8 -> 15.6 | Improved |
| pinball uncertainty instead of MSE+Drop dropout variance (ablation) | LSTM model · Gaze360 test (uncertainty correlation) | Spearman uncertainty correlation: 0.31 -> 0.45 | Improved |
| pinball uncertainty instead of crop augmentation variance (ablation) | LSTM model · Gaze360 test (uncertainty correlation) | Spearman uncertainty correlation: 0.37 -> 0.45 | Improved |
| pinball uncertainty instead of crop augmentation (ablation) | LSTM model · Gaze360 test (front facing) | mean angular error (deg): 11.2 -> 11.1 | Improved |
| train on Gaze360 instead of Columbia | Static model · Cross-dataset test on MPIIFaceGaze | mean angular error (deg): 12.3 -> 12.1 | Improved |
| train on Gaze360 instead of RT-GENE | Static model · Cross-dataset test on Columbia | mean angular error (deg): 24.2 -> 9.0 | Improved |
| add self-supervised domain adaptation (discriminator and flip-symmetry loss) (ablation) | Gaze360 Static · Cross-dataset test on Columbia | mean angular error (deg): 9.0 -> 8.1 | Improved |
| add self-supervised domain adaptation (ablation) | Gaze360 Static · Cross-dataset test on MPIIFaceGaze | mean angular error (deg): 12.1 -> 9.9 | Improved |
| add self-supervised domain adaptation (ablation) | Gaze360 Static · Cross-dataset test on RT-GENE | mean angular error (deg): 23.4 -> 21.9 | Improved |
Later papers that compared with this method
4 later accepted papers report Gaze360 as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- GFIE: A Dataset and Baseline for Gaze-Following From 2D to 3D in Indoor Environments (CVPR 2023 · better on 4 of 4 shared results)
- Look Both Ways: Self-Supervising Driver Gaze Estimation and Road Scene Saliency (ECCV 2022 · better on 4 of 4 shared results)
- Dynamic 3D Gaze From Afar: Deep Gaze Estimation From Temporal Eye-Head-Body Coordination (CVPR 2022 · better on 1 of 1 shared result)
- Dual Attention Guided Gaze Target Detection in the Wild (CVPR 2021 · better on 3 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
- Dynamic 3D Gaze From Afar: Deep Gaze Estimation From Temporal Eye-Head-Body Coordination (CVPR 2022 · similarity 0.91)
- Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels (CVPR 2025 · similarity 0.90)
- 3DGazeNet: Generalizing Gaze Estimation with Weak Supervision from Synthetic Views (ECCV 2024 · similarity 0.89)
- OmniGaze: Reward-inspired Generalizable Gaze Estimation in the Wild (NeurIPS 2025 · similarity 0.89)
- GA3CE: Unconstrained 3D Gaze Estimation with Gaze-Aware 3D Context Encoding (CVPR 2025 · similarity 0.89)
- MTGS: A Novel Framework for Multi-Person Temporal Gaze Following and Social Gaze Prediction (NeurIPS 2024 · similarity 0.89)
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.
