MVHumanNet
收藏资源简介:
MVHumanNet是由香港中文大学(深圳)创建的一个大规模多视角人类动作捕捉数据集,包含4500个不同身份的人和9000种日常服装。该数据集通过48个高分辨率摄像头捕捉,涵盖了60000个动作序列和6.45亿帧视频,具有丰富的标注信息,如人体掩码、相机参数、2D/3D关键点等。MVHumanNet旨在解决3D视觉领域中人类中心任务的数据稀缺问题,支持多种下游任务,如动作识别、神经辐射场重建和文本驱动的人体图像生成。
MVHumanNet is a large-scale multi-view human motion capture dataset developed by The Chinese University of Hong Kong, Shenzhen. It encompasses 4500 distinct individuals and 9000 types of daily clothing. Captured via 48 high-resolution cameras, this dataset covers 60000 motion sequences and 645 million video frames, with rich annotation information including human masks, camera parameters, 2D/3D keypoints, and so forth. MVHumanNet aims to address the data scarcity issue in human-centric tasks within the 3D vision field, and supports a variety of downstream tasks such as action recognition, neural radiance field (NeRF) reconstruction, and text-driven human image generation.

- 1MVHumanNet: A Large-scale Dataset of Multi-view Daily Dressing Human Captures香港中文大学(深圳) · 2023年



