RekaDaily-10k-raw

RekaDaily-10k-raw是一个大规模原始自我中心视频数据集，专为第一人称视觉、具身AI、多模态学习和AI代理感知研究而设计。该数据集包含397,171个第一人称日常生活视频，总计约7,834小时，通过Reka的Claru数据收集市场收集。视频由付费收集者使用头戴式摄像头和手持手机在多样化的真实世界环境中录制，包括家庭、工作场所、商业环境和日常活动场景。数据集保留原始收集格式，未经重新编码、剪辑或内容过滤，仅进行基本完整性检查，提供用户视角的真实视觉体验。它包含丰富的元数据，如活动分类、收集项目、光照条件、视频时长、帧信息和环境标识符。RekaDaily-10k-raw支持的研究任务包括自我中心活动理解、人-物交互分析、视频-语言学习、具身AI、机器人感知和多模态基础模型训练。该数据集适用于需要真实世界视觉理解和第一人称环境感知的智能代理和系统的开发。RekaDaily-10k-raw is a large-scale raw egocentric video dataset designed for research on first-person vision, embodied AI, multimodal learning, and AI agent perception. The dataset contains 397,171 first-person daily-life videos totaling approximately 7,834 hours, collected through Reka’s Claru data collection marketplace. The videos are recorded by paid collectors using head-mounted cameras and handheld phones across diverse real-world environments, including homes, workplaces, commercial environments, and daily activity scenarios. The dataset preserves videos in their original collected format without re-encoding, cutting, or content filtering beyond basic integrity checks, providing authentic visual experiences from the user’s perspective. It includes rich metadata such as activity taxonomy, collection projects, lighting conditions, video duration, frame information, and environmental identifiers. RekaDaily-10k-raw supports research tasks including egocentric activity understanding, human-object interaction analysis, video-language learning, embodied AI, robotics perception, and multimodal foundation model training. The dataset is suitable for developing intelligent agents and systems that require real-world visual understanding and first-person environmental awareness.