OpenEDS2020

OpenEDS2020 是在受控照明下以 100 Hz 的帧速率捕获的眼睛图像序列数据集，使用安装有两个同步的面向眼睛的摄像头的虚拟现实头戴式显示器。该数据集匿名删除参与者的任何个人身份信息，由 80 名不同外表的参与者组成，执行多项凝视引发任务，并分为两个子集：1) 凝视预测数据集，多达 66,560 个序列，包含 550,400 只眼睛- 图像和相应的凝视矢量，旨在促进时空凝视估计和预测方法的研究； 2) 眼部分割数据集，由 200 个以 5 Hz 采样的序列组成，包含多达 29,500 张图像，其中 5% 包含语义分割标签，旨在鼓励使用时间信息将标签传播到连续帧。

OpenEDS2020 is a dataset of eye image sequences captured at a frame rate of 100 Hz under controlled lighting, using a virtual reality head-mounted display equipped with two synchronized eye-facing cameras. All personal identifying information of participants has been anonymized and removed. This dataset consists of 80 participants with diverse appearances who completed multiple gaze-eliciting tasks, and is divided into two subsets:
1) Gaze Prediction Dataset: With up to 66,560 sequences containing 550,400 eye images and corresponding gaze vectors, this subset is designed to promote research on spatiotemporal gaze estimation and prediction methods;
2) Eye Segmentation Dataset: Composed of 200 sequences sampled at 5 Hz, containing up to 29,500 images, 5% of which have semantic segmentation labels, and is intended to encourage the propagation of labels to consecutive frames using temporal information.