Linked Dynamic Graph CNN: Learning through Point Cloud by Linking Hierarchical Features

Zhang，Kuangen1,2; Hao，Ming3; Wang，Jing2; Chen，Xinxing1; Leng，Yuquan1; De Silva，Clarence W.2; Fu，Chenglong1

10.1109/M2VIP49856.2021.9665104

2021

27th International Conference on Mechatronics and Machine Vision in Practice (M2VIP)

978-1-6654-3154-5

2021 27th International Conference on Mechatronics and Machine Vision in Practice, M2VIP 2021

7-12

NOV 26-28, 2021

null,Shanghai,PEOPLES R CHINA

345 E 47TH ST, NEW YORK, NY 10017 USA

IEEE

Learning through a point cloud is attractive be-cause a point cloud contains geometric data and can help robots understand environments in a robust manner. However, a point cloud is sparse, unstructured and unordered, and its accurate recognition by a traditional convolutional neural network (CNN) or a recurrent neural network (RNN) is difficult. Hence, the present paper proposes a linked dynamic graph CNN (LDGCNN) to directly classify and segment a point cloud. The present work removes the transformation network, links hierarchical features from dynamic graphs, freezes the feature extractor, and retrains the classifier in order to increase the performance of LDGCNN. Also, the paper describes LDGCNN using theoretical analysis and visualization. The LDGCNN is able to classify a point cloud consisting of 1024 three dimensional points without normal vectors in the ModelNet40 dataset, achieving 92.9% accuracy, which is better than the state-of-art. The LDGCNN also achieves state-of-the-art segmentation performance in the ShapeNet dataset.

Point cloud compression Training Solid modeling Recurrent neural networks Semantics Object segmentation Feature extraction

第一 ; 通讯

英语

CPCI-S ; EI

National Natural Science Foundation of China[U1913205];

Automation & Control Systems ; Computer Science ; Engineering

Automation & Control Systems ; Computer Science, Artificial Intelligence ; Engineering, Multidisciplinary

WOS:000783817900002

20220811682717

Classification (of information) ; Convolutional neural networks ; Graph neural networks

Information Theory and Signal Processing:716.1 ; Artificial Intelligence:723.4 ; Information Sources and Analysis:903.1

2-s2.0-85124797161

Scopus

1.Southern University of Science and Technology,Shenzhen Key Laboratory of Biomimetic Robotics and Intelligent Systems and Guangdong Provincial,Key Laboratory of Human-Augmentation and Rehabilitation Robotics in Universities,Department of Mechanical and Energy Engineering,Shenzhen,518055,China
2.University of British Columbia,Department of Mechanical Engineering,Vancouver,V6T1Z4,Canada
3.Tsinghua University,Department of Mechanical Engineering,Beijing,100084,China

Zhang，Kuangen,Hao，Ming,Wang，Jing,et al. Linked Dynamic Graph CNN: Learning through Point Cloud by Linking Hierarchical Features[C]. 345 E 47TH ST, NEW YORK, NY 10017 USA:IEEE,2021:7-12.