MoviCam
收藏资源简介:
MoviCam是一个创新的评估基准,用于在复杂场景中跟踪运动相机下的人类运动。该数据集是首个包含详细场景几何信息、全局人类运动和移动相机轨迹的数据集,提供了精确的3D人体姿态和形状,以及人-场景接触标签。数据集包含7个序列,其中5个序列在非平面地面上进行,2个序列在平坦地面上进行。每个序列都展示了不同个体与场景的交互,并使用不同的移动相机轨迹进行捕获。这些交互包括行走、跳跃、伸展和蹲坐等。数据集共包含约22,000张图像,为人体运动跟踪研究提供了重要的评估资源。
MoviCam is an innovative evaluation benchmark for human motion tracking under moving cameras in complex scenes. This dataset is the first of its kind to include detailed scene geometry information, global human motion, and moving camera trajectories, while providing accurate 3D human poses and shapes as well as human-scene contact labels. The dataset consists of 7 sequences, 5 of which are conducted on non-planar ground and 2 on flat ground. Each sequence showcases interactions between different individuals and the scene, captured using distinct moving camera trajectories. These interactions include walking, jumping, stretching, squatting, and other similar movements. In total, the dataset contains approximately 22,000 images, serving as a critical evaluation resource for human motion tracking research.
PhyDynPose数据集概述
数据集基本信息
- 名称:MoviCam
- 关联方法:PhysDynPose
- 发布会议:CVPRW 2025
- 论文标题:Physics-based Human Pose Estimation from a Single Moving RGB Camera
- 论文链接:https://openaccess.thecvf.com/content/CVPR2025W/RHOBIN/papers/Aytekin_Physics-based_Human_Pose_Estimation_from_a_Single_Moving_RGB_Camera_CVPRW_2025_paper.pdf
数据集状态
- 当前状态:待发布(代码和数据集将很快发布)
研究领域
- 主要应用:基于物理的人体姿态估计
- 数据采集方式:单移动RGB相机

- 1Physics-based Human Pose Estimation from a Single Moving RGB Camera马克斯·普朗克计算机科学研究所 · 2025年



