RoadSocial
收藏资源简介:
RoadSocial数据集是由印度海得拉巴国际信息学院创建的一个大规模、多样化的视频问答(VideoQA)数据集,专为从社交视频叙述中理解通用道路事件而设计。该数据集从涵盖1400万帧和41.4万条社交评论的社交媒体视频中获得,包含13.2K个视频、674个标签和26万个高质量的问题回答对。数据集利用文本和视频大型语言模型(LLM)的半自动标注框架,在12个具有挑战性的任务中生成全面的问题回答对。
The RoadSocial Dataset is a large-scale, diverse video question answering (VideoQA) dataset created by the International Institute of Information Technology, Hyderabad, India, specifically designed for understanding general road events from social video narratives. Derived from social media videos spanning 14 million frames and 414,000 social comments, it contains 13.2K videos, 674 labels, and 260,000 high-quality question-answer pairs. The dataset adopts a semi-automatic annotation framework leveraging text and video large language models (LLMs) to generate comprehensive question-answer pairs across 12 challenging tasks.




