Long Video Dataset

该数据集包含三个视频，每个视频都记录了一个对象超过7000帧的画面。目标对象在视频中有较长的运动轨迹和突然的外观变化，这导致了显著的表示漂移。此外，该数据集被用于评估在长视频条件下，所提出持续学习方法的有效性。数据规模为三个视频，任务为视频对象分割。This dataset includes three videos, each capturing footage of a single object spanning over 7000 frames. The target object features long motion trajectories and abrupt appearance changes within the videos, which induces significant representation drift. Moreover, this dataset is utilized to evaluate the effectiveness of the proposed continual learning method under long-video conditions. The associated task for this dataset is video object segmentation.