Relation-Shape Convolutional Neural Network for Point Cloud Analysis

By Yongcheng Liu, Bin Fan, Shiming Xiang, Chunhong Pan. Computer Vision and Pattern Recognition, 2019

1,032 citations recorded by the source corpus.

Point cloud analysis is very challenging, as the shape implied in irregular points is difficult to capture. In this paper, we propose RS-CNN, namely, Relation-Shape Convolutional Neural Network, which extends regular grid CNN to irregular configuration for point cloud analysis. The key to RS-CNN is learning from relation, i.e., the geometric topology constraint among points. Specifically, the convolutional weight for local point set is forced to learn a high-level relation expression from predefined geometric priors, between a sampled point from this point set and the others. In this way, an inductive local representation with explicit reasoning about the spatial layout of points can be obtained, which leads to much shape awareness and robustness. With this convolution as a basic operator, RS-CNN, a hierarchical architecture can be developed to achieve contextual shape-aware learning for point cloud analysis. Extensive experiments on challenging benchmarks across three tasks verify RS-CNN achieves the state of the arts.

Subfields: Computational Mechanics, Computer Graphics and Computer-Aided Design, Geology

View Paper: https://www.semanticscholar.org/paper/a58715797b61588cbd020b9a98292a98f8483420

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

1. Jiwen Lu
2. Li Jiang
3. Cheng Wang
4. Saifullahi Aminu Bello
5. Shangshu Yu
6. Yulan Guo
7. Jie Zhou
8. Jiageng Mao

In Computer Graphics and Computer-Aided Design

1. Jiwen Lu
2. Cheng Wang
3. Jing Shao
4. Yu Qiao
5. Stefano Rosa
6. Saifullahi Aminu Bello
7. Shangshu Yu
8. Jia Deng

In Geology

1. Jiwen Lu
2. Li Jiang
3. Cheng Wang
4. Saifullahi Aminu Bello
5. Shangshu Yu
6. Hongsheng Li
7. Jie Zhou
8. Jiageng Mao

Citations in the corpus

• PCT: Point cloud transformer (2020, neutral): “2 RS-CNN [20] 86.”
• Geometric Back-Projection Network for Point Cloud Classification (2019, methodology): “We are also higher than RS-CNN [23], which uses post-processing of a ten votes evaluation arrangement during testing.”
• Stratified Transformer for 3D Point Cloud Segmentation (2022, neutral): “…[50] 82.3 85.2 SubSparseCNN [15] 83.3 86.0 PointCNN [22] 84.6 86.1 PointConv [52] 82.8 85.7 Point2Sequence [23] - 85.2 PVCNN [27] - 86.2 RS-CNN [25] 84.0 86.2 KPConv [41] 85.0 86.2 InterpCNN [30] 84.0 86.3 DensePoint [24] 84.2 86.4 PAConv [54] 84 Point Transformer by a large margin in category…”
• PointASNL: Robust Point Clouds Processing Using Nonlocal Neural Networks With Adaptive Sampling (2020, methodology): “In particular, when the input contains only 64 points, PNL can even help to improve the accuracy of our model, from 73.9% to 85.2%, which largely exceeds the current state-of-the-art RS-CNN [22] (about 75%).”
• Review: deep learning on 3D point clouds (2020, methodology): “R-S CNN [38] Uniform sampling radius-nn MLP”
• Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework (2022, methodology): “Different from the models with sophisticated local geometric extractors (e.g., DeepGCNs (Li et al., 2019), RSCNN (Liu et al., 2019b), RPNet (Ran et al., 2021).)”
• CrossPoint: Self-Supervised Cross-Modal Contrastive Learning for 3D Point Cloud Understanding (2022, background): “Since then, numerous advancements have been made in point cloud based tasks such as 3D object classification [27, 29, 39, 49, 55, 62, 70], 3D object detection [27, 32, 37, 69] and 3D point cloud synthesis [1, 57].”
• 2DPASS: 2D Priors Assisted Semantic Segmentation on LiDAR Point Clouds (2022, background): “Later on, many studies design point-wise MLP [25,26], adaptive weight [27, 28] and pseudo grid [29, 30] based methods to extract local features of point clouds or exploit nonlocal operators [31–33] to learn long distance dependency.”
• Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline (2021, methodology): “…nonlocally geometric fea-tures” (Yan et al., 2020), that a projection-base method “often demands a huge number of views for decent per-formance” (Liu et al., 2019c), and that projection-based methods often “fine-tune a pre-trained image-based architecture for accurate recognition” (Liu et…”
• Point Transformer V2: Grouped Vector Attention and Partition-based Pooling (2022, background): “For the four encoding stages, the feature dimensions are [96, 192, 384, 384], and the corresponding attention groups are [12, 24, 48, 48].”
• SoftGroup for 3D Instance Segmentation on Point Clouds (2022, background): “Continuous convolution methods [23, 40, 44, 45] learn the kernels which are associated to the spatial distribution of local points.”
• PF-Net: Point Fractal Network for 3D Point Cloud Completion (2020, background): “However, multi-view and voxel-based representation[18, 7, 34, 11, 13, 37] leads to unnecessarily voluminous and limits the output resolution of voxels.”
• Global-Local Bidirectional Reasoning for Unsupervised Representation Learning of 3D Point Clouds (2020, methodology): “To show our method can be used for various point cloud networks, we consider two baseline models: PointNet++ [38] and Relation-Shape CNN (RSCNN) [33].”
• DeepVCP: An End-to-End Deep Neural Network for Point Cloud Registration (2019, background): “We are also considering to try rotation invariant 3D descriptors [6, 16, 23] in the future.”