A-CNN: Annularly Convolutional Neural Networks on Point Clouds

CVPR 2019  ·  Artem Komarichev, Zichun Zhong, Jing Hua ·

Analyzing the geometric and semantic properties of 3D point clouds through the deep networks is still challenging due to the irregularity and sparsity of samplings of their geometric structures. This paper presents a new method to define and compute convolution directly on 3D point clouds by the proposed annular convolution. This new convolution operator can better capture the local neighborhood geometry of each point by specifying the (regular and dilated) ring-shaped structures and directions in the computation. It can adapt to the geometric variability and scalability at the signal processing level. We apply it to the developed hierarchical neural networks for object classification, part segmentation, and semantic segmentation in large-scale scenes. The extensive experiments and comparisons demonstrate that our approach outperforms the state-of-the-art methods on a variety of standard benchmark datasets (e.g., ModelNet10, ModelNet40, ShapeNet-part, S3DIS, and ScanNet).

PDF Abstract CVPR 2019 PDF CVPR 2019 Abstract

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
3D Point Cloud Classification ModelNet40 A-CNN Overall Accuracy 92.6 # 49
Semantic Segmentation S3DIS PointCNN Mean IoU 65.4 # 27
oAcc 88.1 # 16
Semantic Segmentation S3DIS 3P-RNN Mean IoU 56.3 # 39
oAcc 86.9 # 21
Semantic Segmentation S3DIS A-CNN Mean IoU 62.9 # 31
oAcc 87.3 # 19
Semantic Segmentation S3DIS SPGraph Mean IoU 62.1 # 32
oAcc 85.5 # 23
Semantic Segmentation S3DIS PointNet Mean IoU 47.6 # 41
oAcc 78.5 # 26