THUD++
收藏资源简介:
THUD++是由清华大学深圳国际研究生院创建的大规模动态室内场景数据集,专为移动机器人设计。该数据集包含13个动态场景,结合了真实世界和合成数据,涵盖了超过90,000帧RGB-D图像和2000万个2D/3D边界框。数据集的创建过程结合了真实机器人平台和Unity3D模拟平台,确保了数据的真实性和多样性。THUD++主要应用于移动机器人的动态场景理解任务,如3D物体检测、语义分割、轨迹预测和导航,旨在解决复杂动态环境中的机器人感知和导航问题。
THUD++ is a large-scale dynamic indoor scene dataset developed by the Tsinghua Shenzhen International Graduate School, specifically tailored for mobile robots. This dataset comprises 13 dynamic scenarios, integrating real-world and synthetic data, and encompasses over 90,000 RGB-D image frames and 20 million 2D/3D bounding boxes. The dataset was constructed using both real robotic platforms and Unity3D simulation platforms, ensuring the authenticity and diversity of the data. THUD++ is primarily applied to dynamic scene understanding tasks for mobile robots, such as 3D object detection, semantic segmentation, trajectory prediction and navigation, aiming to address the challenges of robot perception and navigation in complex dynamic environments.




