MSGCNN: Multi-scale Graph Convolutional Neural Network for Point Cloud Segmentation

Point cloud has emerged as a scalable and flexible geometric representation for 3D data. Graph convolutional neural networks (GCNNs) have shown superior performance and robustness in point cloud processing with structure-awareness and permutation invariance. However, naive graph convolution networks are limited in point cloud segmentation tasks especially in the border areas of multiple segmentation instances due to the lack of multi-scale feature extraction ability. In this paper, we propose a novel multi-scale graph convolutional neural network (MSGCNN) to allow multi-scale feature learning for fine-grained point cloud segmentation. The proposed geometrical interpretable multi-scale point cloud processing framework is able to considerately enlarge the graph filters receptive fields and exploit discriminative multi-scale structure-aware point features for the superior segmentation performance against naive graph convolution networks especially in border area. Experimental results for part segmentation task on ShapeNet datasets show that MSGCNN achieves competitive performance with state-of-the-arts. In comparison to naive graph convolution networks, MSGCNN is shown to obtain better visual quality in the border area. We further validate that our model is robust to data point missing and noise perturbation with the learned multi-scale structure-aware point features.