EgoOops
收藏资源简介:
EgoOops数据集由京都大学创建,包含50个第一人称视角视频,涵盖五个不同的程序性任务领域,包括电路、颜色混合实验、离子反应实验、玩具积木和纸板工艺。数据集总时长为6.8小时,每个视频配有相应的程序性文本。数据集的创建过程包括任务选择、程序性文本准备、视频录制和标注,涵盖视频与文本的对齐、错误标签及其描述。EgoOops旨在解决从第一人称视角视频中检测错误动作的问题,适用于智能档案系统,能够检测工人的错误并提供反馈。
The EgoOops dataset was created by Kyoto University. It contains 50 first-person perspective videos covering five distinct procedural task domains, including electronic circuit assembly, color mixing experiments, ion reaction experiments, toy block building, and cardboard crafting. The total runtime of the dataset is 6.8 hours, and each video is paired with its corresponding procedural text. The dataset creation process includes task selection, procedural text preparation, video recording and annotation, covering video-text alignment, error label annotation and their corresponding descriptions. EgoOops aims to solve the problem of erroneous action detection from first-person perspective videos, and is applicable to intelligent archive systems that can detect workers' mistakes and provide feedback.
EgoOops 数据集概述
数据集描述
EgoOops 是一个视频-语言数据集,记录了包括错误动作在内的程序性活动。该数据集专注于从第一人称视角视频中检测错误动作,并结合程序性文本进行分析。
数据集特点
- 视频来源: 包含50个第一人称视角视频,涵盖五个程序性领域。
- 文本信息: 每个视频对应程序性文本,用于描述视频内容。
- 标注类型: 数据集包含三种类型的标注:
- 视频-文本对齐
- 错误标签
- 错误描述,解释每个片段中的错误
相关链接
作者信息
- Yuto Haneji<sup>1</sup>
- Taichi Nishimura<sup>2</sup>
- Hirotaka Kameko<sup>1</sup>
- Keisuke Shirai<sup>1</sup>
- Tomoya Yoshida<sup>1</sup>
- Keiya Kajimura<sup>1</sup>
- Koki Yamamoto<sup>1</sup>
- Taiyu Cui<sup>1</sup>
- Tomohiro Nishimoto<sup>1</sup>
- Shinsuke Mori<sup>1</sup>
机构:
- <sup>1</sup> Kyoto University
- <sup>2</sup> LY Corporation




