ScanNet v2
收藏资源简介:
ScanNet 是一个 RGB-D 视频数据集,包含 1500 多次扫描中的 250 万个视图,并使用 3D 相机姿势、表面重建和实例级语义分割进行注释。为了收集这些数据,我们设计了一个易于使用且可扩展的 RGB-D 捕获系统,其中包括自动表面重建和众包语义注释。我们表明,使用这些数据有助于在几个 3D 场景理解任务上实现最先进的性能,包括 3D 对象分类、语义体素标记和 CAD 模型检索。
ScanNet is an RGB-D video dataset consisting of 2.5 million views across more than 1500 scans, annotated with 3D camera poses, surface reconstruction outputs, and instance-level semantic segmentation. To curate this dataset, we developed an easy-to-use and scalable RGB-D capture system that integrates automatic surface reconstruction and crowdsourced semantic annotation. We demonstrate that utilizing this dataset enables state-of-the-art performance on a range of 3D scene understanding tasks, including 3D object classification, semantic voxel labeling, and CAD model retrieval.




