Motion expressions Video Segmentation (MeViS)
收藏资源简介:
MeViS是由南洋理工大学创建的大型视频分割数据集,专注于基于运动表达的视频对象分割。该数据集包含2006个视频,总计8171个对象,提供了28570个运动表达来指示这些对象。创建过程中,MeViS强调视频内容中的多对象共存和运动,排除了仅包含静态特征的孤立对象视频。数据集的应用领域主要集中在复杂视频场景中,利用运动表达作为主要线索进行对象分割和识别,旨在解决现有数据集在处理视频内容中运动属性方面的不足。
MeViS is a large-scale video segmentation dataset developed by Nanyang Technological University, focusing on motion expression-based video object segmentation. This dataset contains 2006 videos with a total of 8171 object instances, and provides 28570 motion expressions to indicate these objects. During its creation, MeViS emphasizes the coexistence of multiple objects and motion in video content, and excludes isolated single-object videos that only contain static features. Its application scenarios mainly focus on complex video scenes, where motion expressions are used as the primary cue for object segmentation and recognition, aiming to address the shortcomings of existing datasets in handling the motion attributes of video content.




