Kinetics-GEBD
收藏资源简介:
Kinetics-GEBD数据集是由新加坡国立大学与Facebook AI共同创建的一个新型视频分析基准。该数据集包含54691个视频片段,专注于检测和分割视频中的通用事件边界,无需预定义的事件类别。数据集的创建基于认知科学的研究,旨在模拟人类自然地将视频内容分割成有意义的片段的能力。Kinetics-GEBD数据集不仅涵盖了广泛的视频领域,还采用了开放词汇而非预定义的分类体系,使得数据集能够捕捉到人类感知的多样性。此数据集的应用领域包括视频编辑、摘要、关键帧选择和亮点检测,旨在推动长格式视频理解的发展。
The Kinetics-GEBD dataset is a novel video analysis benchmark jointly developed by the National University of Singapore and Facebook AI. It comprises 54,691 video clips, focusing on detecting and segmenting general event boundaries in videos without pre-defined event categories. Grounded in cognitive science research, the dataset is designed to simulate the human capacity to naturally segment video content into semantically meaningful segments. Covering a wide spectrum of video domains and adopting an open-vocabulary framework rather than a pre-defined classification system, the Kinetics-GEBD dataset enables the capture of the diversity of human perceptual experience. Its application areas include video editing, summarization, keyframe selection, and highlight detection, with the ultimate goal of advancing the development of long-form video understanding.




