Point clouds provide a flexible geometric representation suitable for countless applications in computer graphics; they also comprise the raw output of most 3D data acquisition devices. While hand-designed features on point clouds have long been proposed in graphics and vision, however, the recent overwhelming success of convolutional neural networks (CNNs) for image analysis suggests the value of adapting insight from CNN to the point cloud world. Point clouds inherently lack topological information so designing a model to recover topology can enrich the representation power of point clouds. To this end, we propose a new neural network module dubbed EdgeConv suitable for CNN-based high-level tasks on point clouds including classification and segmentation. EdgeConv acts on graphs dynamically computed in each layer of the network. It is differentiable and can be plugged into existing architectures. Compared to existing modules operating in extrinsic space or treating each point independently, EdgeConv has several appealing properties: It incorporates local neighborhood information; it can be stacked applied to learn global shape properties; and in multi-layer systems affinity in feature space captures semantic characteristics over potentially long distances in the original embedding. We show the performance of our model on standard benchmarks including ModelNet40, ShapeNetPart, and S3DIS.
核心贡献 · Key contributions
提出 EdgeConv，一种新颖的点云神经网络模块，在保持置换不变性的同时捕获局部几何结构。 Proposes EdgeConv, a novel neural network module for point clouds that captures local geometric structure while maintaining permutation invariance.
引入动态图更新，每层基于特征空间重新计算 k 近邻图，实现非局部语义分组。 Introduces dynamic graph updates, where the k-NN graph is recomputed in each layer based on feature space, enabling non-local semantic grouping.
证明 EdgeConv 可集成到 PointNet 等现有架构中，提升分类和分割性能。 Demonstrates that EdgeConv can be integrated into existing architectures like PointNet to improve performance on classification and segmentation.
在 ModelNet40、ShapeNetPart 和 S3DIS 基准上取得最先进结果，且计算高效。 Achieves state-of-the-art results on ModelNet40, ShapeNetPart, and S3DIS benchmarks with efficient computation.
提供理论分析，表明 EdgeConv 将 PointNet 和图 CNN 作为特例包含在内。 Provides theoretical analysis showing EdgeConv subsumes PointNet and graph CNNs as special cases.
发布代码以促进可重复性和未来研究。 Releases code to facilitate reproducibility and future research.
局限 · Limitations
k 近邻中 k 值过大时性能下降，因欧氏距离无法近似测地距离。 Performance degrades with large k in k-NN due to Euclidean distance failing to approximate geodesic distance.
当点数少于 512 时，模型鲁棒性显著下降。 Model robustness drops significantly when fewer than 512 points are used.