GroundLie360
收藏资源简介:
GroundLie360数据集是由新加坡国立大学等机构创建的,包含了2000多个经过事实核查的视频,这些视频包含虚假类型和定位注释。数据集提供了对虚假内容的统一基准,包括假标题/假演讲、时间编辑、计算机生成图像、矛盾内容和未支持内容等五种虚假类型。数据集旨在解决在线虚假视频带来的社会风险,通过细粒度的多模态定位,帮助识别和解释虚假内容,为可解释的多模态虚假信息检测奠定了基础。
The GroundLie360 dataset was created by institutions including the National University of Singapore, and contains over 2,000 fact-checked videos annotated with deception types and localization information. It provides a unified benchmark for misinformation, covering five categories of deceptive content: fake headlines/fake speeches, temporal editing, computer-generated imagery (CGI), contradictory content, and unsupported content. This dataset aims to address the social risks posed by online deceptive videos, and facilitates the identification and interpretation of misinformation through fine-grained multimodal localization, laying a foundation for explainable multimodal misinformation detection.




