Physics Anomaly Detection (Phys-AD)
收藏资源简介:
Physics Anomaly Detection (Phys-AD)数据集是由上海科技大学等机构创建的第一个大规模、现实世界、基于物理的视频数据集,用于工业异常检测。该数据集包含22个类别的49个对象与机器人臂和电机交互的6400多个视频,展示了47种异常类型。数据集通过真实的机器人臂和电机收集,捕捉了机器人臂或电机与工业对象之间的交互过程,旨在推动视觉推理和物理知识结合的异常检测算法的发展。
The Physics Anomaly Detection (Phys-AD) dataset is the first large-scale, real-world, physics-based video dataset for industrial anomaly detection, created by institutions including ShanghaiTech University. It contains over 6,400 videos of 49 objects interacting with robot arms and motors across 22 categories, showcasing 47 types of anomalies. Collected via real robot arms and motors, the dataset captures the interaction processes between robot arms or motors and industrial objects, aiming to promote the development of anomaly detection algorithms that integrate visual reasoning and physical knowledge.




