Research map / 3D vision / Human pose estimation
ETH-XGaze: A Large Scale Dataset for Gaze Estimation under Extreme Head Pose and Gaze Variation
Xucong Zhang, Seonwook Park, Thabo Beeler, Derek Bradley, Siyu Tang, Otmar Hilliges
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 |
|---|---|---|---|
| train with all 16 lighting conditions instead of full-lighting only (ablation) | ResNet-50 baseline · ETH-XGaze test (TE) | gaze error (deg): 8.5 -> 7.8 | Improved |
| train at 224x224 instead of 112x112 (tested at matching size) (ablation) | ResNet-50 baseline · ETH-XGaze | gaze error (deg): 5.4 -> 4.5 | Improved |
| train and test at 448x448 instead of 224x224 (ablation) | ResNet-50 baseline · ETH-XGaze | gaze error (deg): 4.5 -> 4.2 | Improved |
| test at a different resolution than training (224 train, 448 test) (ablation) | ResNet-50 baseline · ETH-XGaze | gaze error (deg): 4.5 -> 42.1 | Hurt |
| train with 112x112 and test at 448x448 instead of matched (ablation) | ResNet-50 baseline · ETH-XGaze | gaze error (deg): 5.4 -> 37.2 | Hurt |
| constrain training head poses to +-40 degrees (ablation) | ResNet-50 baseline · ETH-XGaze test (TE) | gaze error (deg): 34.6% performance decrease | Hurt |
| constrain training gaze directions to +-40 degrees (ablation) | ResNet-50 baseline · ETH-XGaze test (TE) | gaze error (deg): 82.1% performance decrease | Hurt |
| constrain both head poses and gaze directions to +-40 degrees (ablation) | ResNet-50 baseline · ETH-XGaze test (TE) | gaze error (deg): 206.4% performance decrease | Hurt |
| increase number of training participants (ablation) | ResNet-50 baseline · ETH-XGaze | gaze error (deg) | Improved |
| add 200 personal calibration samples to fine-tune | ResNet-50 baseline · ETH-XGaze person-specific | gaze error (deg) | Improved |
| train on ETH-XGaze instead of GazeCapture, tested on Gaze360 | ResNet-50 baseline · Gaze360 test | gaze error (deg): 30.2 -> 27.3 | Improved |
| train on GazeCapture instead of ETH-XGaze, tested on MPIIGaze | ResNet-50 baseline · MPIIGaze test | gaze error (deg): 4.5 -> 7.5 | Hurt |
Closest papers by meaning
- MPIIGaze: Real-World Dataset and Deep Appearance-Based Gaze Estimation (TPAMI 2017 · similarity 0.91)
- Towards High Performance Low Complexity Calibration in Appearance Based Gaze Estimation (TPAMI 2022 · similarity 0.89)
- UVAGaze: Unsupervised 1-to-2 Views Adaptation for Gaze Estimation (AAAI 2024 · similarity 0.87)
- OmniGaze: Reward-inspired Generalizable Gaze Estimation in the Wild (NeurIPS 2025 · similarity 0.87)
- Suppressing Uncertainty in Gaze Estimation (AAAI 2024 · similarity 0.86)
- Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels (CVPR 2025 · 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.
