ManiPose

ManiPose是由上海交通大学计算机科学与工程系创建的综合性基准数据集，专注于姿态感知物体操作研究。该数据集包含3036个真实世界扫描的刚性物体和100个关节物体，涵盖59个类别，提供几何一致且面向操作的6D姿态标签。数据集的创建过程涉及从多个源整合对象网格并标准化姿态标签，以支持大规模的姿态变化操作任务。ManiPose的应用领域广泛，旨在通过提供丰富的姿态信息，推动机器人操作技能的发展，特别是在日常场景和非结构化环境中的应用。

ManiPose is a comprehensive benchmark dataset created by the Department of Computer Science and Engineering at Shanghai Jiao Tong University, focusing on pose-aware object manipulation research. This dataset contains 3036 real-world scanned rigid objects and 100 articulated objects, covering 59 categories, and provides geometrically consistent and manipulation-oriented 6D pose annotations. The dataset creation process involves integrating object meshes from multiple sources and standardizing pose annotations to support large-scale pose-variant manipulation tasks. ManiPose has broad application scenarios, aiming to advance the development of robotic manipulation skills by providing rich pose information, particularly for applications in daily scenarios and unstructured environments.