FriendsQA
收藏资源简介:
FriendsQA是一个大规模的深度视频理解数据集,由武汉大学创建,基于著名情景喜剧《老友记》构建。该数据集包含44.6K个问题,均匀分布在14个细粒度主题上,平均每集视频长度为1,358秒。数据集的创建过程采用了基于大型语言模型的多智能体协作框架,自动生成并筛选高质量问题,确保了问题的多样性和平衡性。FriendsQA主要用于评估视频问答模型在复杂故事情节理解方面的能力,旨在解决现有数据集在深度视频理解任务中的不足。
FriendsQA is a large-scale deep video understanding dataset created by Wuhan University, constructed based on the famous sitcom Friends. This dataset contains 44.6K questions evenly distributed across 14 fine-grained topics, with an average video length of 1,358 seconds per episode. The dataset was developed using a large language model-based multi-agent collaboration framework, which automatically generates and filters high-quality questions to ensure the diversity and balance of the questions. FriendsQA is primarily used to evaluate the capability of video question answering models in comprehending complex storylines, aiming to address the limitations of existing datasets in deep video understanding tasks.




