Dynamic Graph CNN for Learning on Point Clouds

Abstract

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.

References in corpus (8)

• Semi-Supervised Classification with Graph Convolutional Networks
• Deep Convolutional Networks on Graph-Structured Data
• Deep Continuous Fusion for Multi-Sensor 3D Object Detection
• PCPNET: Learning Local Shape Properties from Raw Point Clouds
• Feature Denoising for Improving Adversarial Robustness
• Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling
• Generating 3D faces using Convolutional Mesh Autoencoders
• Self-supervised Learning of Dense Shape Correspondence

Cited by in corpus (116)

• RPM-Net: Robust Point Matching using Learned Features
• DeepGCNs: Making GCNs Go as Deep as CNNs
• Fake News Detection on Social Media using Geometric Deep Learning
• Linked Dynamic Graph CNN: Learning on Point Cloud via Linking Hierarchical Features
• Learning representations of irregular particle-detector geometry with distance-weighted graph networks
• PointCNN: Convolution On $\mathcal{X}$-Transformed Points
• Deep Closest Point: Learning Representations for Point Cloud Registration
• DeepGCNs: Can GCNs Go as Deep as CNNs?
• PointConv: Deep Convolutional Networks on 3D Point Clouds
• Relation-Shape Convolutional Neural Network for Point Cloud Analysis
• RGCNN: Regularized Graph CNN for Point Cloud Segmentation
• Graph Neural Networks for Particle Reconstruction in High Energy Physics detectors
• Applications and Techniques for Fast Machine Learning in Science
• Adversarial Attack and Defense on Point Sets
• How much real data do we actually need: Analyzing object detection performance using synthetic and real data
• Domain Adaptation on Point Clouds via Geometry-Aware Implicits
• 3D-BEVIS: Bird's-Eye-View Instance Segmentation
• GaitSet: Regarding Gait as a Set for Cross-View Gait Recognition
• PCRNet: Point Cloud Registration Network using PointNet Encoding
• PointASNL: Robust Point Clouds Processing using Nonlocal Neural Networks with Adaptive Sampling
• Embedding Hard Physical Constraints in Neural Network Coarse-Graining of 3D Turbulence
• Iterative Transformer Network for 3D Point Cloud
• Patch-based Progressive 3D Point Set Upsampling
• Point2Sequence: Learning the Shape Representation of 3D Point Clouds with an Attention-based Sequence to Sequence Network
• Pose from Shape: Deep Pose Estimation for Arbitrary 3D Objects
• Dual-Primal Graph Convolutional Networks
• Dynamic Graph Convolutional Networks Using the Tensor M-Product
• PCN: Point Completion Network
• PVNet: A Joint Convolutional Network of Point Cloud and Multi-View for 3D Shape Recognition
• PointNetKL: Deep Inference for GICP Covariance Estimation in Bathymetric SLAM
• GAPNet: Graph Attention based Point Neural Network for Exploiting Local Feature of Point Cloud
• Shared Data and Algorithms for Deep Learning in Fundamental Physics
• 3D Point Cloud Descriptors in Hand-crafted and Deep Learning Age: State-of-the-Art
• Deep Learning for LiDAR Point Clouds in Autonomous Driving: A Review
• LPD-Net: 3D Point Cloud Learning for Large-Scale Place Recognition and Environment Analysis
• FlowNet3D: Learning Scene Flow in 3D Point Clouds
• Mitigating the Hubness Problem for Zero-Shot Learning of 3D Objects
• Generating 3D Adversarial Point Clouds
• Distilling Knowledge from Graph Convolutional Networks
• MortonNet: Self-Supervised Learning of Local Features in 3D Point Clouds
• PointGrow: Autoregressively Learned Point Cloud Generation with Self-Attention
• Cascaded Refinement Network for Point Cloud Completion
• PDE-GCN: Novel Architectures for Graph Neural Networks Motivated by Partial Differential Equations
• DUP-Net: Denoiser and Upsampler Network for 3D Adversarial Point Clouds Defense
• PeerNets: Exploiting Peer Wisdom Against Adversarial Attacks
• H3DNet: 3D Object Detection Using Hybrid Geometric Primitives
• Spherical Kernel for Efficient Graph Convolution on 3D Point Clouds
• Identification of Patterns in Cosmic-Ray Arrival Directions using Dynamic Graph Convolutional Neural Networks
• PointNetLK: Robust & Efficient Point Cloud Registration using PointNet
• NeuralSampler: Euclidean Point Cloud Auto-Encoder and Sampler
• Graph Neural Networks for IceCube Signal Classification
• Optimized Skeleton-based Action Recognition via Sparsified Graph Regression
• Modeling Local Geometric Structure of 3D Point Clouds using Geo-CNN
• Learning 2D-3D Correspondences To Solve The Blind Perspective-n-Point Problem
• 3D-Rotation-Equivariant Quaternion Neural Networks
• JSNet: Joint Instance and Semantic Segmentation of 3D Point Clouds
• iPool -- Information-based Pooling in Hierarchical Graph Neural Networks
• One Framework to Register Them All: PointNet Encoding for Point Cloud Alignment
• X-Vision: An augmented vision tool with real-time sensing ability in tagged environments
• DiffGCN: Graph Convolutional Networks via Differential Operators and Algebraic Multigrid Pooling
• Learning Video Representations from Correspondence Proposals
• Real-time Progressive 3D Semantic Segmentation for Indoor Scene
• NormalNet: Learning-based Normal Filtering for Mesh Denoising
• SRINet: Learning Strictly Rotation-Invariant Representations for Point Cloud Classification and Segmentation
• Convolutional Neural Networks on non-uniform geometrical signals using Euclidean spectral transformation
• Point2Node: Correlation Learning of Dynamic-Node for Point Cloud Feature Modeling
• Progressive Point Cloud Deconvolution Generation Network
• Unsupervised Feature Learning for Point Cloud by Contrasting and Clustering With Graph Convolutional Neural Network
• Adaptive Hierarchical Down-Sampling for Point Cloud Classification
• PointCloud Saliency Maps
• Graph-Revised Convolutional Network
• Geometry Sharing Network for 3D Point Cloud Classification and Segmentation
• Every Node Counts: Self-Ensembling Graph Convolutional Networks for Semi-Supervised Learning
• Point-Based Multi-View Stereo Network
• Latent-Space Laplacian Pyramids for Adversarial Representation Learning with 3D Point Clouds
• Point Clouds Learning with Attention-based Graph Convolution Networks
• Attention to the strengths of physical interactions: Transformer and graph-based event classification for particle physics experiments
• Point-to-Pose Voting based Hand Pose Estimation using Residual Permutation Equivariant Layer
• Exploring Hypergraph Representation on Face Anti-spoofing Beyond 2D Attacks
• Multi-Resolution Graph Neural Network for Large-Scale Pointcloud Segmentation
• Exploring Structure-Adaptive Graph Learning for Robust Semi-Supervised Classification
• Robust 6D Object Pose Estimation by Learning RGB-D Features
• Graph Neural Processes: Towards Bayesian Graph Neural Networks
• 3DCFS: Fast and Robust Joint 3D Semantic-Instance Segmentation via Coupled Feature Selection
• PVRNet: Point-View Relation Neural Network for 3D Shape Recognition
• Endowing Deep 3D Models with Rotation Invariance Based on Principal Component Analysis
• Robust Deep Graph Based Learning for Binary Classification
• Multi-view PointNet for 3D Scene Understanding
• Unsupervised Learning of Intrinsic Structural Representation Points
• PointHop++: A Lightweight Learning Model on Point Sets for 3D Classification
• Spherical Transformer: Adapting Spherical Signal to CNNs
• Learning Embedding of 3D models with Quadric Loss
• Occlusion Resistant Object Rotation Regression from Point Cloud Segments
• Facial Action Unit Intensity Estimation via Semantic Correspondence Learning with Dynamic Graph Convolution
• PointNLM: Point Nonlocal-Means for vegetation segmentation based on middle echo point clouds
• Going Deeper with Lean Point Networks
• Knowledge Embedding Based Graph Convolutional Network
• A Deeper Look at 3D Shape Classifiers
• Local-Area-Learning Network: Meaningful Local Areas for Efficient Point Cloud Analysis
• URSA: A Neural Network for Unordered Point Clouds Using Constellations
• Go Wider: An Efficient Neural Network for Point Cloud Analysis via Group Convolutions
• POIRot: A rotation invariant omni-directional pointnet
• Modelling Diffuse Subcellular Protein Structures as Dynamic Social Networks
• Fast Hierarchical Neural Network for Feature Learning on Point Cloud
• Large-scale 3D point cloud representations via graph inception networks with applications to autonomous driving
• Irregular Convolutional Auto-Encoder on Point Clouds
• BoxNet: A Deep Learning Method for 2D Bounding Box Estimation from Bird's-Eye View Point Cloud
• Justlookup: One Millisecond Deep Feature Extraction for Point Clouds By Lookup Tables
• DeepSemanticHPPC: Hypothesis-based Planning over Uncertain Semantic Point Clouds
• Cross-Level Cross-Scale Cross-Attention Network for Point Cloud Representation
• 3D Siamese Voxel-to-BEV Tracker for Sparse Point Clouds
• Surrogate Modelling for Injection Molding Processes using Machine Learning
• Zero-shot Learning of 3D Point Cloud Objects
• Picasso: A CUDA-based Library for Deep Learning over 3D Meshes
• MNEW: Multi-domain Neighborhood Embedding and Weighting for Sparse Point Clouds Segmentation
• Verifiability and Predictability: Interpreting Utilities of Network Architectures for Point Cloud Processing