Atypical Video Dataset
收藏资源简介:
本文介绍了一个名为“异常视频数据集”的新视频数据集,包含7,818个视频片段,涵盖了科幻、动画、异常、无意和超现实等多种类型的非典型数据。数据集由多个来源组成,包括YouTube电影预告片、动画预告片、舞台表演视频、现有的异常检测数据集以及合成场景数据集。该数据集旨在支持开放世界学习研究,并用于研究如何利用非典型数据来增强模型的泛化能力。
This paper presents a novel video dataset dubbed the "Anomalous Video Dataset", which contains 7,818 video clips covering a range of atypical data types including science fiction, animation, anomalous, unintentional, and surreal scenarios. The dataset is compiled from multiple sources, namely YouTube movie trailers, animation trailers, stage performance videos, existing anomaly detection datasets, and synthetic scene datasets. This dataset is intended to support open-world learning research, and is utilized to investigate how to leverage atypical data to enhance the generalization ability of models.




