合成人类活动识别数据集

本数据集由普林斯顿大学创建，旨在通过3D渲染技术生成大量用于训练人类活动识别模型的视频数据。数据集包含150个视频，涵盖三种常见动作：走路、挥手和坐下。创建过程中，结合了MoVi数据集的关节位置数据和人体渲染包，生成的视频具有多样化的视角和相机运动。该数据集主要用于解决人类活动识别领域中视频数据标注和存储的高成本问题，通过合成数据提高模型对真实视频的泛化能力。

This dataset was developed by Princeton University to generate extensive video data for training human activity recognition models using 3D rendering technologies. It consists of 150 videos covering three common human actions: walking, waving, and sitting. During the creation process, joint position data from the MoVi dataset and human rendering packages were integrated to produce videos with diverse viewpoints and camera movements. This dataset is primarily intended to resolve the high-cost issues of video data annotation and storage in the human activity recognition domain, and to enhance the generalization ability of models towards real-world videos via synthetic data.