Human-M3
收藏资源简介:
Human-M3数据集是由清华大学自动化系创建,专注于户外场景中的三维人体姿态估计。该数据集包含89642条有效的人体姿态记录,不仅包括多视角RGB视频,还包含相应的点云数据,以提供更全面的环境信息。数据集的创建过程中,采用了基于多模态数据输入的算法来生成地面实况标注,解决了户外多人场景中人体定位不准确和匹配模糊的问题。Human-M3数据集适用于评估户外多人三维人体姿态估计算法的性能,特别是在解决复杂环境下的姿态估计问题方面具有重要价值。
The Human-M3 dataset, developed by the Department of Automation, Tsinghua University, focuses on 3D human pose estimation in outdoor scenarios. This dataset contains 89,642 valid human pose records, including not only multi-view RGB videos but also corresponding point cloud data to provide more comprehensive environmental information. During the dataset construction process, algorithms based on multimodal data inputs were adopted to generate ground truth annotations, addressing the issues of inaccurate human localization and ambiguous matching in outdoor multi-person scenarios. The Human-M3 dataset is suitable for evaluating the performance of outdoor multi-person 3D human pose estimation algorithms, and holds significant value particularly in solving pose estimation challenges in complex environments.

- 1Human-M3: A Multi-view Multi-modal Dataset for 3D Human Pose Estimation in Outdoor Scenes清华大学自动化系 · 2023年



