migHug/AmbientEye

AmbientEye 是一个大规模眼动追踪数据集（n = 2,606,225），专门在户外自然阳光下收集，不使用任何主动红外照明器。该数据集旨在基准测试在真实世界户外使用中出现的、多变且不受控制的近红外（NIR）辐照条件下的瞳孔分割方法。为了获得高质量的瞳孔注释，每个会话的第一帧瞳孔区域内手动放置一个点作为SAM2的提示。在第二阶段，人工注释者审查并细化分割结果的每一帧以验证其准确性。当预测的掩码与瞳孔边界对齐不佳时，注释者通过手动标记瞳孔边界点来拟合椭圆以修正注释。总共有2,518,693帧（占96.6%）获得了瞳孔注释。数据集包括35名参与者的70个会话（每人2个会话），总帧数为2,606,225，条件分为sunfacing（面向太阳，较高NIR）和sunoccluded（背向太阳，较低NIR）。使用Xreal Air 2眼镜配备两个OV6211单色红外摄像头（分辨率400×400像素，帧率120fps）进行录制，所有NIR辐照完全来自环境阳光。数据集结构包括原始视频文件、提取的帧（需通过脚本生成）、经人工审核的SAM2分割轮廓、元数据（如校准日志和录制时间）、参与者人口统计信息、NIR辐照测量数据等。AmbientEye is a large-scale eye-tracking dataset (n = 2,606,225) collected exclusively outdoors under natural ambient sunlight — without any active infrared illuminator. It is designed to benchmark pupil segmentation methods under varying, uncontrolled NIR irradiance conditions that arise in real-world outdoor use. To obtain high-quality pupil annotations, a single point is manually placed within the pupil region of the first frame of each session and provided as a prompt for SAM2. In the second stage, human annotators review and refine every frame of the segmentation result to validate its accuracy. When the predicted mask does not align well with the pupil boundary, annotators correct the annotation by manually marking pupil boundary points to fit an ellipse to the pupil. In total, pupil annotations were obtained for 2,518,693 out of 2,606,225 frames (96.6%). The dataset includes 70 sessions (2 per participant) from 35 participants, with a total of 2,606,225 frames, and conditions are categorized as sunfacing (facing the sun, higher NIR) and sunoccluded (facing away from the sun, lower NIR). Recording was performed using Xreal Air 2 Glasses equipped with two OV6211 monochrome IR cameras (resolution 400×400 px, frame rate 120 fps), with all NIR irradiance originating entirely from ambient sunlight. The dataset structure comprises raw video files, extracted frames (to be generated via script), human-reviewed SAM2 segmentation contours, metadata (such as calibration logs and recording timing), participant demographics, NIR irradiance measurements, and more.