UBnormal
收藏资源简介:
UBnormal是由布加勒斯特大学创建的视频异常检测数据集,包含29个虚拟场景和236,902个视频帧。数据集通过Cinema4D软件生成,包含多种正常和异常事件,如行走、战斗、跳舞等,旨在通过像素级标注支持全监督学习方法。UBnormal特别之处在于训练和测试集中的异常类型不重叠,确保了开放集问题的特性。该数据集适用于评估和比较开放集和封闭集模型,并已证明能提升现有异常检测框架在Avenue和ShanghaiTech数据集上的性能。
UBnormal is a video anomaly detection dataset created by the University of Bucharest, which includes 29 virtual scenarios and 236,902 video frames. Generated using Cinema4D software, the dataset covers a variety of normal and abnormal events such as walking, fighting, dancing and others. It features pixel-level annotations to support fully supervised learning methods. A distinct feature of UBnormal is that the anomaly types in the training and test sets do not overlap, which ensures its open-set problem characteristics. This dataset is applicable for evaluating and comparing open-set and closed-set models, and has been proven to improve the performance of existing anomaly detection frameworks on the Avenue and ShanghaiTech datasets.

- 1UBnormal: New Benchmark for Supervised Open-Set Video Anomaly Detection布加勒斯特大学 · 2023年



