PointCloud-C

3D感知，尤其是点云分类和部分分割，取得了实质性进展。然而，在实际部署中，由于场景复杂、传感器不准确和处理不精确，点云损坏是不可避免的。在这项工作中，我们贡献了 ModelNet-C 和 ShapeNet-C，旨在严格基准测试和分析各种现实世界损坏下的点云鲁棒性。

3D perception, particularly point cloud classification and part segmentation, has achieved substantial progress. However, point cloud corruption is inevitable in real-world deployment due to complex scenes, inaccurate sensors, and imprecise processing. In this work, we introduce ModelNet-C and ShapeNet-C, which are designed as rigorous benchmarks to test and analyze the robustness of point cloud models under various real-world corruptions.

搜集汇总

背景与挑战

背景概述

PointCloud-C是一个用于3D感知的点云分类和部分分割的数据集，旨在测试和分析各种现实世界损坏下的点云鲁棒性。该数据集由南洋理工大学于2020年6月发布，相关论文可在arXiv上获取。