TutorialVQA
收藏arXiv2020-05-31 更新2024-06-21 收录
下载链接：
https://github.com/acolas1/TutorialVQAData
下载链接
链接失效反馈官方服务：
资源简介：
TutorialVQA数据集由佛罗里达大学和Adobe研究共同创建，专注于教程视频中的问题回答任务。该数据集包含约6,195个手动收集的三元组，包括视频、问题和答案跨度，主要来源于屏幕录制教程视频，涉及图像编辑软件的教学内容。数据集的创建过程涉及从76个教程视频中手动分割出408个视频段，并由亚马逊Mechanical Turk工作者生成相应的问题-答案对。TutorialVQA数据集的应用领域主要在于提高非事实性问题的视频问答技术，解决复杂任务的多步骤答案需求。
The TutorialVQA dataset, co-created by the University of Florida and Adobe Research, focuses on the question answering task in tutorial videos. This dataset contains approximately 6,195 manually collected triplets including videos, questions, and answer spans, which are primarily sourced from screen-recorded tutorial videos covering instructional content for image editing software. The dataset creation process involves manually segmenting 408 video clips from 76 tutorial videos, and generating corresponding question-answer pairs via Amazon Mechanical Turk workers. The main application scenarios of the TutorialVQA dataset lie in advancing video question answering technologies for non-factual questions, and addressing the demand for multi-step answers to complex tasks.
提供机构：
佛罗里达大学, Adobe研究
创建时间：
2019-12-03
