EYEDIAP

缺乏用于从 RGB 和 RGB-D 数据评估凝视估计任务的通用基准是区分文献中发现的许多提议算法的优缺点的严重限制。
EYEDIAP 数据集旨在满足对来自远程 RGB 和 RGB-D（标准视觉和深度）相机的注视估计的标准数据库的需求。记录方法的设计使我们系统地包括并隔离了影响远程注视估计算法的大多数变量：
头部姿势变化。
人物变异。
环境和传感条件的变化。
目标类型：屏幕或 3D 对象。
提供了一些预定义的基准来以独立的方式评估这些方面中的每一个，并对数据进行预处理以提取和提供补充观察（例如头部姿势）。

The lack of a universal benchmark for evaluating gaze estimation tasks using RGB and RGB-D data poses a critical limitation to distinguishing the strengths and weaknesses of numerous proposed algorithms reported in the literature.
The EYEDIAP dataset aims to address the demand for a standardized database for gaze estimation from remote RGB and RGB-D (standard visual and depth) cameras.
The recording methodology was designed to systematically include and isolate most variables that affect remote gaze estimation algorithms:
- Head pose variations.
- Participant variability.
- Variations in environmental and sensing conditions.
- Target types: screens or 3D objects.
A set of predefined benchmarks are provided to evaluate each of these aspects independently, and the data is preprocessed to extract and provide supplementary observations (e.g., head pose).