多粒度视频-文本预训练数据集
收藏资源简介:
该数据集是通过Granularity EXpansion (GEX)方法从单粒度视频-文本数据集扩展而来的多粒度数据集,旨在解决视频-文本学习任务中的跨模态对齐问题。数据集通过视频和文本的集成与压缩操作生成,能够反映真实世界中视频和文本的多粒度特性。创建过程简单且可扩展,无需额外的人工标注。该数据集主要应用于视频-文本对齐、视频分类和视频问答等任务,旨在提升模型在长视频理解任务中的表现。
This dataset is a multi-granularity dataset expanded from single-granularity video-text datasets via the Granularity EXpansion (GEX) method, aiming to address the cross-modal alignment problem in video-text learning tasks. It is generated through integration and compression operations on videos and texts, and can reflect the multi-granularity characteristics of videos and texts in real-world scenarios. Its construction process is simple and scalable, with no need for additional manual annotations. This dataset is mainly applied to tasks such as video-text alignment, video classification, and video question answering, aiming to improve the performance of models in long video understanding tasks.




