APSIPA Transactions on Signal and Information Processing > Vol 12 > Issue 1

A Tiny Machine Learning Model for Point Cloud Object Classification

Min Zhang, University of Southern California, USA, zhan980@alumni.usc.edu , Jintang Xue, University of Southern California, USA, Pranav Kadam, University of Southern California, USA, Hardik Prajapati, University of Southern California, USA, Shan Liu, Tencent Media Lab, Tencent America, USA, C.-C. Jay Kuo, University of Southern California, USA
 
Suggested Citation
Min Zhang, Jintang Xue, Pranav Kadam, Hardik Prajapati, Shan Liu and C.-C. Jay Kuo (2023), "A Tiny Machine Learning Model for Point Cloud Object Classification", APSIPA Transactions on Signal and Information Processing: Vol. 12: No. 1, e35. http://dx.doi.org/10.1561/116.00000114

Publication Date: 16 Aug 2023
© 2023 M. Zhang, J. Xue, P. Kadam, H. Prajapati, S. Liu and C.-C. Jay Kuo
 
Subjects
 
Keywords
Point cloudsobject classificationtiny learning modelgreen learningPointHop
 

Share

Open Access

This is published under the terms of CC BY-NC.

Downloaded: 537 times

In this article:
Introduction 
Related Work 
Green-PointHop Method 
Experiments 
Conclusion and Future Work 
References 

Abstract

The design of a tiny machine learning model which can be deployed in mobile and edge devices for point cloud object classification is investigated in this work. To achieve this objective, we replace the multi-scale representation of a point cloud object with a single-scale representation for complexity reduction, and exploit rich 3D geometric information of a point cloud object for performance improvement. The proposed solution is named Green-PointHop due to its low computational complexity. We evaluate the performance of Green-PointHop on ModelNet40 and ScanObjectNN two datasets. Green-PointHop has a model size of 64 K parameters. It demands 2.3 M floating-point operations (FLOPs) to classify a ModelNet40 object of 1024 down-sampled points. Its classification performance gaps against the state-of-the-art DGCNN method are 3% and 7% for ModelNet40 and ScanObjectNN, respectively. On the other hand, the model size and inference complexity of DGCNN are 42X and 1203X of those of Green-PointHop, respectively.

DOI:10.1561/116.00000114