SSCBench
收藏资源简介:
SSCBench是一个大规模的单目3D语义场景补全基准,专注于街道视图。该数据集由纽约大学创建,旨在推动自主系统中的3D场景理解技术。SSCBench整合了多个广泛使用的汽车数据集,如KITTI-360、nuScenes和Waymo,以提供多样化的街道场景。数据集包含约67,000帧,是现有数据集SemanticKITTI的约8倍大小,覆盖了六个城市的多样地理环境。SSCBench不仅关注单目输入的SSC方法,还使用三目和点云输入来比较不同视图和传感器方法的性能。此外,SSCBench统一了不同数据集的语义标签,简化了跨领域泛化测试,并计划持续整合新的汽车数据集和SSC算法,以进一步推动该领域的发展。
SSCBench is a large-scale monocular 3D semantic scene completion benchmark focused on street-view scenarios. Developed by New York University (NYU), this dataset aims to advance 3D scene understanding technologies for autonomous systems. SSCBench integrates multiple widely adopted automotive datasets, including KITTI-360, nuScenes, and Waymo, to provide diverse street-level scenes. It contains approximately 67,000 frames, which is about 8 times the size of the existing SemanticKITTI dataset, and covers diverse geographic environments across six cities. SSCBench not only focuses on SSC methods with monocular inputs, but also utilizes trinocular and point cloud inputs to compare the performance of different view-based and sensor-based approaches. Furthermore, SSCBench unifies the semantic labels across different datasets, simplifying cross-domain generalization testing, and plans to continuously integrate new automotive datasets and SSC algorithms to further advance the development of this field.




