DEXROBOT/DexCanvas

DexCanvas是一个大规模的混合型数据集，用于机器人手-物体交互研究，结合了真实人类演示和物理验证的模拟数据。该数据集包含约3000万帧多视角RGB-D图像，总时长约70小时。其中真实演示数据约0.7小时，模拟数据是真实数据的100倍。数据集涵盖了21种基于Cutkosky分类法的操作类型，包含30种物体（几何原始物体和YCB物体）。每个帧包括多视角彩色和深度图像、手部姿态、物体状态和注释。需要注意的是，版本v0.1中不包含接触力数据，这些数据将在未来版本中提供。

DexCanvas is a large-scale hybrid dataset for robotic hand-object interaction research, combining real human demonstrations with physics-validated simulation data. The dataset contains about 30 million multi-view RGB-D frames, with a total duration of about 70 hours. It includes approximately 0.7 hours of real human demonstration data and 100 times expansion of simulated data from real data. The dataset covers 21 types of manipulations based on the Cutkosky grasp taxonomy, and includes 30 objects (geometric primitives and YCB objects). Each frame includes multi-view color and depth images, hand poses, object states, and annotations. It is worth noting that contact force data is not included in version v0.1, which will be provided in future releases.