ManipGPT Dataset
收藏资源简介:
ManipGPT Dataset是由北京大学计算机科学学院创建的一个用于机器人操作的视觉可操作性数据集。该数据集包含13350张图像,涵盖30个对象类别,分为训练集和测试集。数据集结合了模拟和真实世界的图像,旨在提高机器人操作的鲁棒性和适应性。数据集的创建过程包括从SAPIEN环境生成模拟图像,以及从多个公开数据集和自采集图像中获取真实世界图像。该数据集主要用于解决机器人操作中的视觉可操作性问题,特别是在复杂和动态环境中的操作任务。
The ManipGPT Dataset is a visual manipulability dataset for robotic manipulation created by the School of Computer Science, Peking University. It contains 13,350 images spanning 30 object categories, and is split into training and test sets. The dataset combines both simulated and real-world images, aiming to improve the robustness and adaptability of robotic manipulation systems. Its creation process includes generating simulated images from the SAPIEN simulation environment, as well as collecting real-world images from multiple public datasets and self-collected samples. This dataset is primarily designed to address visual manipulability challenges in robotic manipulation, especially for manipulation tasks in complex and dynamic environments.




