GSPN: Generative Shape Proposal Network for 3D Instance Segmentation in Point Cloud
GSPN: Generative Shape Proposal Network for 3D Instance Segmentation in Point Cloud
复制标题
DOI：
10.1109/cvpr.2019.00407
复制
发表时间：
2018-12
期刊：
影响因子：
--
通讯作者：
L. Yi;Wang Zhao;He Wang;Minhyuk Sung;L. Guibas
中科院分区：
文献类型：
--
作者：
L. Yi;Wang Zhao;He Wang;Minhyuk Sung;L. Guibas
我们介绍了一种新的3D对象的建议方法命名为生成形状建议网络（GSPN），例如在点云数据分割。而不是把对象的建议作为一个直接的边界框回归问题，我们采取了分析合成策略，并通过重建从噪声观测的形状在场景中生成的建议。我们将GSPN到一个新的3D实例分割框架命名为基于区域的PointNet（R-PointNet），它允许灵活的建议细化和实例分割生成。我们在多个3D实例分割任务上实现了最先进的性能。GSPN的成功很大程度上来自于它在对象建议过程中强调几何理解，大大减少了低对象性的建议。
We introduce a novel 3D object proposal approach named Generative Shape Proposal Network (GSPN) for instance segmentation in point cloud data. Instead of treating object proposal as a direct bounding box regression problem, we take an analysis-by-synthesis strategy and generate proposals by reconstructing shapes from noisy observations in a scene. We incorporate GSPN into a novel 3D instance segmentation framework named Region-based PointNet (R-PointNet) which allows flexible proposal refinement and instance segmentation generation. We achieve state-of-the-art performance on several 3D instance segmentation tasks. The success of GSPN largely comes from its emphasis on geometric understandings during object proposal, greatly reducing proposals with low objectness.