Kinetic-GEB+

Kinetic-GEB+数据集由新加坡国立大学创建，包含超过170,000个边界，这些边界与描述12,000多个视频中通用事件状态变化的标题相关联。数据集涵盖了Kinetic-400数据集的所有400个类别，旨在通过状态变化推动视频理解的精细化、鲁棒性和类人化。数据集创建过程中，采用了严格的标注指南和格式，确保了数据的高质量和一致性。该数据集的应用领域包括AI助手机器人等，旨在通过视觉状态变化和自然语言标题的理解，实现对实时、瞬发事件的无提示辅助。

The Kinetic-GEB+ dataset was developed by the National University of Singapore. It contains over 170,000 boundaries, which are paired with captions that describe state changes of general events across more than 12,000 videos. The dataset covers all 400 categories from the Kinetic-400 dataset, and is designed to advance the refinement, robustness and human-like capability of video understanding by focusing on state changes. During the dataset construction process, strict annotation guidelines and formatting rules were adopted to ensure high data quality and consistency. Its application domains include AI assistant robots and other fields, with the goal of enabling prompt-free assistance for real-time, transient events by comprehending visual state changes and natural language captions.