MultiVENT 2.0
收藏资源简介:
MultiVENT 2.0是由约翰斯·霍普金斯大学创建的大规模多语言事件中心视频检索基准数据集。该数据集包含超过218,000条新闻视频和3,906个查询,涵盖特定世界事件。数据集内容包括专业编辑的新闻广播和手机拍摄的原始第一人称视频,主要涵盖阿拉伯语、中文、英语、韩语、俄语和西班牙语六种语言。数据集的创建过程包括从InternVid等来源扩展视频集合,并通过专业语言学家的注释生成查询。MultiVENT 2.0旨在解决现有视频检索数据集在多样性和规模上的不足,特别是在多模态和多语言检索任务中的应用。
MultiVENT 2.0 is a large-scale multilingual event-centric video retrieval benchmark dataset created by Johns Hopkins University. This dataset contains over 218,000 news videos and 3,906 queries focused on specific global events. It includes both professionally edited news broadcasts and raw first-person footage captured on mobile devices, covering six core languages: Arabic, Chinese, English, Korean, Russian, and Spanish. The development of MultiVENT 2.0 involved expanding the video corpus from sources such as InternVid, and generating queries via annotations from professional linguists. This benchmark aims to address the limitations of existing video retrieval datasets in terms of diversity and scale, especially for applications in multimodal and multilingual retrieval tasks.




