README.md

metadata

license: mit

MobilePoG: Benchmark for Mobile Point-of-Gaze Calibration.

Dataset Summary

Figure 1: Collection procedure of the MobilePoG dataset.

MobilePoG was introduced in the paper "Pose-Robust Calibration Strategy for Point-of-Gaze Estimation on Mobile Phones" (BMVC 2025). It was collected and curated to support research on point-of-gaze estimation and calibration, with the primary goal of enabling valid evaluation of personalized calibration under varying points-of-gaze and head poses.

Dataset Structure

Static-MobilePoG
    -{subject}_{pose}
      --frames
        ---00000.jpg
        ---00001.jpg
        ...
      --camera.json   # camera parameters
      --face.json     # facial landmarks
      --headpose.json # headposes (pitch, yaw, roll)
      --info.json     # information about subject, pose, dataset split
      --label.json    # point-of-gaze labels
   -{subject}_{pose}
   ...

Dynamic-MobilePoG
    -{subject}_{pog}
      --frames
        ---00000.jpg
        ---00001.jpg
        ...
      --camera.json   # camera parameters
      --face.json     # facial landmarks
      --headpose.json # headposes (pitch, yaw, roll)
      --info.json     # information about subject, point-of-gaze, dataset split
      --label.json    # point-of-gaze labels
    -{subject}_{pog}
    ...

Dataset Examples

Figure 2: Examples of the Static-MobilePoG dataset.

Figure 3: Examples of the Dynamic-MobilePoG dataset.

Citation

Researchers using this dataset are kindly requested to cite the following reference:

@misc{zhao2025poserobustcalibrationstrategypointofgaze,
      title={Pose-Robust Calibration Strategy for Point-of-Gaze Estimation on Mobile Phones}, 
      author={Yujie Zhao and Jiabei Zeng and Shiguang Shan},
      year={2025},
      eprint={2508.10268},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2508.10268}, 
}