IEEE Transactions on Pattern Analysis and Machine Intelligence · 2021 · 68 citations · 37 references
ComputingComputer ScienceComputational GeometryComputer VisionGeometric ModelingGeometry ProcessingGraph-based LearningPoint Cloud ProcessingScene UnderstandingShape AnalysisThree-dimensional GeometryAblation StudiesGraph Convolutional NetworksGraph Neural NetworksMax PoolingPoint Cloud ClassificationPoint Cloud Representation LearningPoint Cloud SegmentationSemantic Information
Point clouds are a popular 3D representation, but their unordered nature makes semantic analysis difficult, and existing methods suffer from performance drops under shift and scale variations. This work proposes 3D graph convolution networks (3D‑GCN) to learn 3D kernels with graph max‑pooling for extracting geometric features across scales. 3D‑GCN learns 3D convolutional kernels and applies graph max‑pooling to aggregate features at multiple scales from unordered point sets. The proposed 3D‑GCN achieves joint shift and scale invariance and demonstrates strong performance on point‑cloud classification and segmentation, as confirmed by ablation studies and visualizations.
Point clouds are among the popular geometry representations in 3D vision. However, unlike 2D images with pixel-wise layouts, such representations containing unordered data points which make the processing and understanding the associated semantic information quite challenging. Although a number of previous works attempt to analyze point clouds and achieve promising performances, their performances would degrade significantly when data variations like shift and scale changes are presented. In this paper, we propose 3D graph convolution networks (3D-GCN), which uniquely learns 3D kernels with graph max-pooling mechanisms for extracting geometric features from point cloud data across different scales. We show that, with the proposed 3D-GCN, satisfactory shift and scale invariance can be jointly achieved. We show that 3D-GCN can be applied to point cloud classification and segmentation tasks, with ablation studies and visualizations verifying the design of 3D-GCN.
37
Deep Residual Learning for Image Recognition
Kaiming He, Xiangyu Zhang, Shaoqing Ren et al. · 2016 · 222.1K citations · Full text
Fully convolutional networks for semantic segmentation
Jonathan Long, Evan Shelhamer, Trevor Darrell · 2015 · 36.9K citations
Guillem Cucurull, Arantxa Casanova, Adriana Romero et al. · arXiv (Cornell University) · 2017 · 8.3K citations · Full text
Dynamic Graph CNN for Learning on Point Clouds
Yue Wang, Yongbin Sun, Ziwei Liu et al. · ACM Transactions on Graphics · 2019 · 6.7K citations · Full text