Research map / 3D vision / Human pose estimation
MPIIGaze: Real-World Dataset and Deep Appearance-Based Gaze Estimation
Xucong Zhang, Yusuke Sugano, Mario Fritz, Andreas Bulling
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 GazeNet instead of the prior state-of-the-art appearance method | GazeNet · MPIIGaze (cross-dataset, train UT Multiview) | mean angular error (degrees): 13.9 -> 10.8 | Improved |
| use GazeNet instead of the prior state-of-the-art method | GazeNet · EYEDIAP (cross-dataset, train UT Multiview) | mean angular error (degrees): 8% on EYEDIAP (0.9 degrees) | Improved |
| use 60x36 input instead of 15x9 pixels (ablation) | GazeNet · MPIIGaze (cross-dataset, train UT Multiview) | mean angular error (degrees): 11.4 -> 10.8 | Improved |
| use manually annotated facial landmarks instead of detected ones | GazeNet · MPIIGaze+ (cross-dataset, train UT Multiview) | mean angular error (degrees): 10.8 -> 9.8 | Improved |
| pretrain on synthetic data then fine-tune on UT Multiview | GazeNet+ · MPIIGaze+ (cross-dataset) | mean angular error (degrees): 9.8 -> 9.1 | Improved |
| restrict training gaze/head-pose distribution to match target (UT subset) | GazeNet+ · MPIIGaze+ (cross-dataset) | mean angular error (degrees): 9.8 -> 8.0 | Improved |
| use synthetic training data with matched gaze range | GazeNet+ · MPIIGaze+ (cross-dataset) | mean angular error (degrees): 8.0 -> 7.3 | Improved |
| add detected pupil centre feature (ablation) | GazeNet+ · MPIIGaze+ (cross-person, leave-one-person-out) | mean angular error (degrees): 5.4 -> 5.2 | No clear effect |
| add manually annotated pupil centre feature (ablation) | GazeNet+ · MPIIGaze+ (cross-person, leave-one-person-out) | mean angular error (degrees): 5.4 -> 5.0 | No clear effect |
| fuse both eyes by averaging 3D gaze vectors instead of averaging single-eye estimates | GazeNet+ · MPIIGaze+ (both eyes) | mean angular error (degrees): 9.8 -> 7.2 | Improved |
| use generic instead of personal 3D face model for head pose (ablation) | GazeNet+ · MPIIGaze+ (cross-dataset, train UT Multiview) | mean angular error (degrees): 9.8 -> 9.7 | No clear effect |
| train and test within the same dataset instead of cross-dataset | GazeNet+ · MPIIGaze+ | mean angular error (degrees): 9.1 -> 5.4 | Improved |
Closest papers by meaning
- Appearance-Based Gaze Estimation With Deep Learning: A Review and Benchmark (TPAMI 2024 · similarity 0.91)
- ETH-XGaze: A Large Scale Dataset for Gaze Estimation under Extreme Head Pose and Gaze Variation (ECCV 2020 · similarity 0.91)
- UVAGaze: Unsupervised 1-to-2 Views Adaptation for Gaze Estimation (AAAI 2024 · similarity 0.89)
- OmniGaze: Reward-inspired Generalizable Gaze Estimation in the Wild (NeurIPS 2025 · similarity 0.88)
- Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels (CVPR 2025 · similarity 0.88)
- GazeGene: Large-scale Synthetic Gaze Dataset with 3D Eyeball Annotations (CVPR 2025 · similarity 0.88)
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.
