Snowpoints: Lightweight neural network for point cloud classification. (December 2022)

• Record Type: Journal Article
• Authors: Xin, Zihao; Wang, Hongyuan; Zhang, Ji
• Is Part Of: Computers & electrical engineering. Volume 104:Part B(2022)
• Publication Date: 2022-12
• Subjects: Point cloud classification -- Lightweight neural network -- Multilayer perceptron -- Deep learning; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2022.108463
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: This paper presents Snowpoints, a new lightweight neural network for point cloud classification that only has about 0.1 million parameters. Advances like a broad range of prior research have used the pooling layers to aggregate the local features, seeking to strengthen the representational power of encoding layers by enhancing the quality of local features extracted throughout its high dimensions. In this work, we focus instead on exploring the relationships among points with deeper layers but fewer parameters and faster speed. On ModelNet40 (Wu et al., 2015), Snowpoints reaches over 92.5% accuracy, which is the first time for an ultra-light network, to the best of our knowledge. Compared to the most famous PointNet++, our Snowpoints trains 2.9 × faster and tests 3.2 × faster with higher accuracy, and we show favorable accuracy and speed trade-off compared to the state-of-the-art models.