UVO (Unidentified Video Objects)
收藏arXiv2021-04-10 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2104.04691v1
下载链接
链接失效反馈官方服务：
资源简介：
UVO是由Facebook AI Research (FAIR)创建的一个大型视频数据集，专注于开放世界中的类别无关对象分割。该数据集包含约1200个视频，每个视频平均有12.29个对象注释，远超其他数据集如DAVIS、YouTube-VOS和YouTube-VIS。UVO的特点在于其密集和详尽的对象掩码注释，适用于多种应用，如对象跟踪和超体素分割，旨在推动超越分类和检测的更复杂视频理解任务的研究。
UVO is a large-scale video dataset developed by Facebook AI Research (FAIR), dedicated to category-agnostic object segmentation in the open world. It consists of around 1,200 videos, with an average of 12.29 object annotations per video—far exceeding the annotation scale of other datasets such as DAVIS, YouTube-VOS and YouTube-VIS. UVO features dense and comprehensive object mask annotations, supporting a wide range of applications including object tracking and superpixel segmentation. It aims to advance research on more complex video understanding tasks that go beyond classification and detection.
提供机构：
Facebook AI Research (FAIR)
创建时间：
2021-04-10
