HVU (Holistic Video Understanding)
收藏资源简介:
HVU 在语义分类中分层组织,重点关注多标签和多任务视频理解作为一个综合问题,包括识别动态场景中的多个语义方面。 HVU 总共包含约 57.2 万个视频,其中包含 900 万个注释,用于训练、验证和测试集,跨越 3142 个标签。 HVU 包含在场景、对象、动作、事件、属性和概念类别上定义的语义方面,这些语义方面自然地捕捉了现实世界的场景。来源:大规模整体视频理解
HVU is hierarchically organized for semantic classification, focusing on multi-label and multi-task video understanding as a unified problem, which involves recognizing multiple semantic aspects in dynamic scenes. In total, HVU comprises approximately 572,000 videos with 9 million annotations across its training, validation, and test splits, spanning 3142 distinct labels. HVU encompasses semantic aspects defined across the categories of scene, object, action, event, attribute, and concept, which naturally capture real-world scenarios. Source: Large-scale Holistic Video Understanding




