TAO

TAO数据集是由卡内基梅隆大学创建的一个大规模多对象跟踪基准，包含2907个高分辨率视频，平均时长半分钟，涵盖833个类别，远超以往的跟踪基准。数据集通过采用自下而上的方法发现大量词汇，要求标注者在视频中任何移动的物体上进行标注，并在事后给予名称。TAO数据集的应用领域广泛，旨在解决开放世界中大型词汇跟踪的挑战，特别是在复杂环境和多变场景下的对象跟踪问题。

The TAO dataset is a large-scale multi-object tracking benchmark created by Carnegie Mellon University. It contains 2907 high-resolution videos with an average duration of half a minute, covering 833 categories, which far exceeds previous tracking benchmarks. The dataset discovers a large vocabulary via a bottom-up approach, requiring annotators to label any moving objects in the videos and assign names to them post-hoc. The TAO dataset has a wide range of application fields, aiming to solve the challenge of large-vocabulary tracking in the open world, especially object tracking problems in complex environments and varying scenarios.