Semantic3d. net

3D点云分类是机器人技术，增强现实和城市规划中的重要任务。机器学习和计算机视觉的最新进展已经证明，复杂的现实任务需要大量的训练数据集来进行分类器训练。同时，到目前为止，还没有用于3D点云分类的数据集，这些数据集在对象表示形式和标记点的数量上都足够丰富。例如，众所周知的奥克兰数据集包含少于200万个标记点。另一个流行的数据集NYU基准测试仅提供室内场景。最后，悉尼城市物体数据集和IQmulus & TerraMobilita竞赛都使用安装在汽车上的3D Velodyne激光雷达，其点密度比静态扫描仪低得多。Vaihingen3D机载基准的计数相同。3D point cloud classification is a critical task in robotics, augmented reality, and urban planning. Recent advances in machine learning and computer vision have demonstrated that complex real-world tasks require large-scale training datasets for classifier training. However, to date, no sufficiently rich dataset for 3D point cloud classification exists in terms of both object representation and the number of labeled points. For instance, the well-known Oakland dataset contains fewer than 2 million labeled points. Another popular dataset, the NYU Benchmark, only provides indoor scenes. Finally, both the Sydney Urban Objects Dataset and the IQmulus & TerraMobilita competition utilize 3D Velodyne LiDAR mounted on vehicles, which has a much lower point density than static scanners. The Vaihingen3D airborne benchmark shares this same limitation.

搜集汇总

背景与挑战

背景概述

Semantic3d.net是一个用于3D点云分类的大规模公开数据集，由苏黎世联邦理工学院于2016年发布，大小为24.0GB。该数据集专注于室外场景，提供丰富的对象表示和高点密度，旨在弥补其他数据集（如奥克兰、NYU和悉尼数据集）在点数量、场景覆盖和点密度方面的局限性，适用于机器人、增强现实和城市规划等任务。