A Rotation-Invariant Framework for Deep Point Cloud Analysis

Abstract

Recently, many deep neural networks were designed to process 3D point clouds, but a common drawback is that rotation invariance is not ensured, leading to poor generalization to arbitrary orientations. In this paper, we introduce a new low-level purely rotation-invariant representation to replace common 3D Cartesian coordinates as the network inputs. Also, we present a network architecture to embed these representations into features, encoding local relations between points and their neighbors, and the global shape structure. To alleviate inevitable global information loss caused by the rotation-invariant representations, we further introduce a region relation convolution to encode local and non-local information. We evaluate our method on multiple point cloud analysis tasks, including shape classification, part segmentation, and shape retrieval. Experimental results show that our method achieves consistent, and also the best performance, on inputs at arbitrary orientations, compared with the state-of-the-arts.

References in corpus (13)

• Adam: A Method for Stochastic Optimization
• TensorFlow: A system for large-scale machine learning
• Improving neural networks by preventing co-adaptation of feature detectors
• ShapeNet: An Information-Rich 3D Model Repository
• DeepICP: An End-to-End Deep Neural Network for 3D Point Cloud Registration
• 3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data
• Monte Carlo Convolution for Learning on Non-Uniformly Sampled Point Clouds
• Spherical CNNs
• A Large Dataset of Object Scans
• PU-Net: Point Cloud Upsampling Network
• GAPNet: Graph Attention based Point Neural Network for Exploiting Local Feature of Point Cloud
• Total Denoising: Unsupervised Learning of 3D Point Cloud Cleaning
• 3D-Rotation-Equivariant Quaternion Neural Networks

Cited by in corpus (13)

• GCN-Denoiser: Mesh Denoising with Graph Convolutional Networks
• Triangle-Net: Towards Robustness in Point Cloud Learning
• PaRot: Patch-Wise Rotation-Invariant Network via Feature Disentanglement and Pose Restoration
• RAPiD-Seg: Range-Aware Pointwise Distance Distribution Networks for 3D LiDAR Segmentation
• PointSmile: Point Self-supervised Learning via Curriculum Mutual Information
• TetraSphere: A Neural Descriptor for O(3)-Invariant Point Cloud Analysis
• Rotation-Adaptive Point Cloud Domain Generalization via Intricate Orientation Learning
• Self-supervised Learning of Rotation-invariant 3D Point Set Features using Transformer and its Self-distillation
• Integrating Sensing and Communication in Cellular Networks via NR Sidelink
• PRIN/SPRIN: On Extracting Point-wise Rotation Invariant Features
• Rotation-Invariant Completion Network
• Point Cloud Upsampling via Disentangled Refinement
• Self-Sampling for Neural Point Cloud Consolidation