Egocentric-100K

Egocentric-100K is a large-scale egocentric video dataset designed for research on first-person vision, embodied AI, and multimodal learning. The dataset contains approximately 100,000 egocentric video samples collected from wearable camera perspectives, capturing real-world human activities, interactions, and environmental contexts. It provides rich visual information from the viewpoint of the user, supporting tasks such as action recognition, activity understanding, video-language learning, human-object interaction analysis, and AI agent perception. By leveraging first-person visual data, Egocentric-100K enables the development and evaluation of intelligent systems that can better understand human behavior and real-world environments from an embodied perspective. The dataset is suitable for applications including robotics, autonomous agents, augmented reality, smart devices, and multimodal foundation model training.Egocentric-100K 是一个面向第一视角视觉（First-Person Vision）、具身智能（Embodied AI）以及多模态学习研究的大规模第一视角视频数据集。该数据集包含约 10 万个第一视角视频样本，通过可穿戴摄像设备采集，从真实人类视角记录日常活动、人与环境交互以及周围场景信息。该数据集提供了丰富的用户视角视觉信息，可支持包括动作识别、活动理解、视频-语言学习、人-物交互分析以及 AI 智能体感知等多种任务。通过利用第一视角视觉数据，Egocentric-100K 能够推动智能系统从具身视角理解人类行为以及真实世界环境，促进更具感知能力和交互能力的 AI 模型开发与评估。该数据集适用于多个应用领域，包括机器人、自动化智能体（Autonomous Agents）、增强现实（AR）、智能设备以及多模态基础模型训练等场景。通过大规模真实世界第一视角数据，Egocentric-100K 为构建能够理解人类活动并适应复杂环境的下一代智能系统提供了重要的数据基础。

搜集汇总

背景与挑战

背景概述

Egocentric-100K是一个大规模的第一人称视频数据集，包含约10万个从可穿戴相机视角采集的真实世界人类活动样本。它支持动作识别、人-物交互分析、视频-语言学习等多模态任务，适用于机器人、增强现实和具身AI研究。