Simitate
收藏资源简介:
Simitate 是一个混合基准测试套件,旨在评估模仿学习的方法。它包含一个包含 1938 个序列的数据集,其中人类在现实环境中进行日常活动。数据集与模拟器的集成紧密结合。提供分辨率为 960×540 的 30Hz 的 RGB 和深度流,以及演示者手部的准确地面真实姿势,以及 120Hz 的 6 DOF 对象。除了数据集,还提供了使用环境的 3D 模型和标记的对象图像。
Simitate is a hybrid benchmark suite intended for evaluating imitation learning approaches. It includes a dataset consisting of 1938 sequences where human participants conduct daily activities in real-world settings. The benchmark features tight integration with simulation platforms. It provides RGB and depth streams at 30Hz with a resolution of 960×540, accurate ground-truth poses of the demonstrator's hands, and 6-degree-of-freedom (6DOF) object poses at 120Hz. In addition to the dataset, 3D models of the experimental environments and annotated object images are also provided.




