MPIIGaze
收藏资源简介:
MPIIGaze数据集由马克斯·普朗克信息学研究所创建,包含213,659张全脸图像及其对应的地面实况注视位置,采集自15名用户在日常使用笔记本电脑时的数据。该数据集通过经验抽样方法确保了连续的注视和头部姿态,以及眼睛外观和光照条件的变化。为了便于跨数据集评估,37,667张图像手动标注了眼睛角、嘴角和瞳孔中心。MPIIGaze数据集不仅在眼睛外观和光照变化方面提供了前所未有的真实性,还涵盖了个体外观的多样性,这在现有数据集中是不存在的。该数据集适用于解决不受约束的注视估计任务,特别是在日常环境和无用户面部外观、环境几何属性及相机图像形成属性假设的情况下。
The MPIIGaze dataset was created by the Max Planck Institute for Informatics, consisting of 213,659 full-face images paired with their corresponding ground-truth gaze locations, collected from 15 users during their daily laptop usage. This dataset ensures continuous gaze and head poses, as well as variations in eye appearance and lighting conditions through an empirical sampling method. To facilitate cross-dataset evaluation, 37,667 of these images were manually annotated with eye corners, mouth corners, and pupil centers. The MPIIGaze dataset not only provides unprecedented realism in terms of eye appearance and lighting variations, but also covers diversity in individual facial appearances, which is absent in existing datasets. This dataset is suitable for addressing unconstrained gaze estimation tasks, particularly in daily environments without assumptions about user facial appearance, environmental geometric properties, and camera image formation attributes.

- 1MPIIGaze: Real-World Dataset and Deep Appearance-Based Gaze Estimation马克斯·普朗克信息学研究所 · 2017年



