MedVidCQA
收藏资源简介:
MedVidCQA数据集是由湖南大学和中国科学院自动化研究所联合创建的,旨在通过自然语言问题定位大型未修剪教学视频中的视觉答案。该数据集包含2710条训练数据,用于视频检索和视觉答案定位两个子任务。数据集的创建过程涉及医学专家手动标注,确保数据质量。MedVidCQA数据集的应用领域主要集中在教学视频的理解和交互,特别是在医疗教育领域,帮助用户通过视频快速找到所需信息。
The MedVidCQA dataset was jointly created by Hunan University and the Institute of Automation, Chinese Academy of Sciences, with the objective of localizing visual answers in large-scale untrimmed instructional videos via natural language queries. This dataset contains 2710 training samples for two subtasks: video retrieval and visual answer localization. The construction of the dataset involved manual annotation performed by medical experts to ensure high data quality. The application scenarios of the MedVidCQA dataset primarily focus on instructional video understanding and interaction, especially in the medical education sector, where it assists users in quickly finding required information from videos.

- 1Learning to Locate Visual Answer in Video Corpus Using Question湖南大学电气与信息工程学院 中国科学院自动化研究所模式识别国家重点实验室 · 2023年



