IU ShareView
收藏资源简介:
IU ShareView数据集由9组两个5-10分钟的第一人称视频组成。每套包含3-4名参与者,他们在六个室内环境之一中进行各种日常活动 (握手,聊天,吃饭等)。每个帧中的每个人都带有一个地面真实边界框和一个唯一的人员ID。为了评估我们关于人物分割的方法,我们用像素级人物注释手动扩充了数据集的子集,总共1,277个带标签的帧包含2,654个带注释的人物实例。
The IU ShareView dataset comprises 9 groups, each containing two first-person videos with a duration ranging from 5 to 10 minutes. Each group involves 3 to 4 participants performing various daily activities such as handshakes, conversations, dining and other routine interactions in one of six indoor environments. Each person appearing in every frame is paired with a ground-truth bounding box and a unique person ID. To evaluate our person segmentation method, we manually augmented a subset of the dataset with pixel-level person annotations. In total, these 1,277 labeled frames contain 2,654 annotated person instances.




