InterAct
收藏资源简介:
InterAct是一个大规模的多模态数据集,捕捉了日常场景中两个人之间的动态、表达性和互动性活动。数据集由241个运动序列组成,每个序列中两个人进行一分钟的表演,涵盖了从日常事件到超自然现象的各种主题。该数据集不仅捕捉了身体动作,还包括面部表情和语音,从而提供了一个全面了解人类行为的视角。InterAct数据集旨在解决目前多模态数据集中缺乏长时间、大范围互动行为捕捉的问题,对于两个人体运动生成、运动估计以及改进现有模型的人类行为基础等方面具有重要作用。
InterAct is a large-scale multimodal dataset that captures dynamic, expressive, and interactive activities between two individuals in daily scenarios. The dataset consists of 241 motion sequences, each containing a 1-minute-long interaction performed by two people, covering a wide range of topics from daily events to supernatural phenomena. This dataset not only captures bodily movements, but also facial expressions and speech, thus providing a comprehensive perspective for understanding human behavior. The InterAct dataset aims to address the current gap in long-duration, large-scale interactive behavior capture within multimodal datasets. It plays a critical role in areas such as two-person motion generation, motion estimation, and improving the human behavior fundamentals of existing models.
- 1通过香港大学 · 2025年



