UrbanScene3D
收藏资源简介:
UrbanScene3D是由深圳大学创建的大型城市场景感知和重建研究数据平台,包含超过12.8万张高分辨率图像,覆盖16个场景,总面积达136平方公里。该数据集不仅包含高精度激光雷达扫描,还有数百个具有不同观测模式的图像集,为设计和评估空中路径规划和3D重建算法提供了全面的基准。此外,该数据集基于Unreal Engine和Airsim模拟器,结合手动标注的独特实例标签,能够生成各种数据,如2D深度图、2D/3D边界框和3D点云/网格分割等。模拟器不仅能够产生多样化的数据,还能让用户在提出的城市环境中模拟汽车或无人机,为未来的研究提供支持。UrbanScene3D旨在解决户外研究中由于缺乏有效设备和大规模尺度带来的数据捕捉难度,以及由于多变的天气和光照条件导致的数据结构化和设计鲁棒采集算法的挑战。
UrbanScene3D is a large-scale urban scene perception and reconstruction research data platform developed by Shenzhen University. It comprises over 128,000 high-resolution images spanning 16 scenarios and covering a total area of 136 square kilometers. This dataset not only includes high-precision LiDAR scans but also hundreds of image sets with distinct observation modes, serving as a comprehensive benchmark for designing and evaluating aerial path planning and 3D reconstruction algorithms. Furthermore, built on the Unreal Engine and Airsim simulators and paired with manually annotated unique instance labels, this dataset supports the generation of diverse data modalities including 2D depth maps, 2D/3D bounding boxes, 3D point cloud and mesh segmentation, among others. The simulators can not only generate varied datasets but also allow users to simulate cars or drones within the proposed urban environment, providing support for future research endeavors. UrbanScene3D aims to resolve the challenges encountered in outdoor research, namely the difficulty of data capture caused by the shortage of effective equipment and large-scale dimensions, as well as the issues of data structuring and the design of robust data collection algorithms arising from volatile weather and lighting conditions.




