EventStream Dataset

EventStream Dataset是由京东集团和同济大学合作创建的机器人抓取检测数据集，包含91个对象。该数据集通过动态视觉传感器（DAVIS）直接从现实世界中采集，采用高频1kHz进行标注，以解决异步事件流标注难题。数据集创建过程中，设计了特殊的标注系统，通过四个LED灯和时空混合粒子滤波（SMP Filter）实现快速而稳健的跟踪。该数据集主要应用于机器人领域，特别是提高机器人在动态环境中的抓取感知能力，解决抓取检测问题。

The EventStream Dataset is a robotic grasping detection dataset co-developed by JD Group and Tongji University, consisting of 91 objects. It is directly collected from the real world via a Dynamic Vision Sensor (DAVIS), and annotated at a high frequency of 1 kHz to tackle the challenge of asynchronous event stream annotation. During the dataset construction process, a dedicated annotation system was designed, which achieves fast and robust tracking through four LED lamps and the Spatio-Temporal Mixed Particle Filter (SMP Filter). This dataset is primarily applied in the robotics field, particularly to improve robots' grasping perception capabilities in dynamic environments and resolve the robotic grasping detection problem.