STAIR Actions
收藏资源简介:
STAIR Actions是由千叶工业大学STAIR实验室创建的大型视频数据集,专注于日常家庭动作识别。该数据集包含100种精细的日常家庭动作类别,共有102,462个视频,每个视频时长主要为5至6秒。数据集的视频主要来源于YouTube或通过众包工作者制作。STAIR Actions旨在支持家庭任务相关的研究,如护理、照顾和安全监控。数据集的创建过程涉及从Wiktionary获取日语基本动词列表,筛选与家庭和办公室常见的动作相关的动词,并定义动作标签。该数据集的应用领域广泛,特别适用于开发和测试用于日常家庭环境中的动作识别模型。
STAIR Actions is a large-scale video dataset developed by the STAIR Laboratory of Chiba Institute of Technology, focusing on daily household action recognition. It comprises 100 fine-grained daily household action categories, with a total of 102,462 videos, each primarily lasting 5 to 6 seconds. The videos in this dataset are mainly sourced from YouTube or created by crowd workers. The core goal of STAIR Actions is to support research related to household tasks such as care, assistance, and safety monitoring. The dataset creation workflow includes retrieving a list of basic Japanese verbs from Wiktionary, filtering out verbs associated with actions commonly encountered in households and offices, and defining the corresponding action labels. This dataset has a wide range of application fields, and is particularly suitable for developing and testing action recognition models deployed in daily household environments.




