Performance of Graph Neural Networks for Point Cloud Applications

被引:0
|
作者
Parikh, Dhruv [1 ]
Zhang, Bingyi [1 ]
Kannan, Rajgopal [2 ]
Prasanna, Viktor [1 ]
Busart, Carl [2 ]
机构
[1] Univ Southern Calif, Los Angeles, CA 90007 USA
[2] DEVCOM US Army Res Lab, Adelphi, MD USA
关键词
Graph neural network; point cloud; k-nearest neighbors; dynamic graph construction; performance profiling;
D O I
10.1109/HPEC58863.2023.10363595
中图分类号
TP3 [计算技术、计算机技术];
学科分类号
0812 ;
摘要
Graph Neural Networks (GNNs) have gained significant momentum recently due to their capability to learn on unstructured graph data. Dynamic GNNs (DGNNs) are the current state-of-the-art for point cloud applications; such applications (viz. autonomous driving) require real-time processing at the edge with tight latency and memory constraints. Conducting performance analysis on such DGNNs, thus, becomes a crucial task to evaluate network suitability. This paper presents a profiling analysis of EdgeConv-based DGNNs applied to point cloud inputs. We assess their inference performance in terms of end-to-end latency and memory consumption on state-of-the-art CPU and GPU platforms. The EdgeConv layer has two stages: (1) dynamic graph generation using k-Nearest Neighbors (kNN) and, (2) node feature updation. The addition of dynamic graph generation via kNN in each (EdgeConv) layer enhances network performance compared to networks that work with the same static graph in each layer; such performance enhancement comes, however, at the added computational cost associated with the dynamic graph generation stage (via kNN algorithm). Understanding its costs is essential for identifying the performance bottleneck and exploring potential avenues for hardware acceleration. To this end, this paper aims to shed light on the performance characteristics of EdgeConv-based DGNNs for point cloud inputs. Our performance analysis on a state-of-the-art EdgeConv network for classification shows that the dynamic graph construction via kNN takes up upwards of 95% of network latency on the GPU and almost 90% on the CPU. Moreover, we propose a quasi-Dynamic Graph Neural Network (qDGNN) that halts dynamic graph updates after a specific depth within the network to significantly reduce the latency on both CPU and GPU whilst matching the original networks inference accuracy.
引用
收藏
页数:7
相关论文
共 50 条
  • [21] Graph neural networks: A review of methods and applications
    Zhou, Jie
    Cui, Ganqu
    Hu, Shengding
    Zhang, Zhengyan
    Yang, Cheng
    Liu, Zhiyuan
    Wang, Lifeng
    Li, Changcheng
    Sun, Maosong
    AI OPEN, 2020, 1 : 57 - 81
  • [22] Exploiting Local Geometry for Feature and Graph Construction for Better 3D Point Cloud Processing with Graph Neural Networks
    Srivastava, Siddharth
    Sharma, Gaurav
    2021 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA 2021), 2021, : 12903 - 12909
  • [23] Graph Neural Networks: Foundation, Frontiers and Applications
    Wu, Lingfei
    Cui, Peng
    Pei, Jian
    Zhao, Liang
    Guo, Xiaojie
    PROCEEDINGS OF THE 29TH ACM SIGKDD CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, KDD 2023, 2023, : 5831 - 5832
  • [24] A survey of graph neural networks and their industrial applications
    Lu, Haoran
    Wang, Lei
    Ma, Xiaoliang
    Cheng, Jun
    Zhou, Mengchu
    Neurocomputing, 2025, 614
  • [25] Towards Efficient Graph Convolutional Networks for Point Cloud Handling
    Li, Yawei
    Chen, He
    Cui, Zhaopeng
    Timofte, Radu
    Pollefeys, Marc
    Chirikjian, Gregory
    Van Gool, Luc
    2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021), 2021, : 3732 - 3742
  • [26] Dynamic graph attention networks for point cloud landslide segmentation
    Wei, Ruilong
    Ye, Chengming
    Ge, Yonggang
    Li, Yao
    Li, Jonathan
    INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION, 2023, 124
  • [27] Generalized Convolutional Neural Networks for Point Cloud Data
    Savchenkov, Aleksandr
    Davis, Andrew
    Zhao, Xuan
    2017 16TH IEEE INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND APPLICATIONS (ICMLA), 2017, : 930 - 935
  • [28] Effective Early Stopping of Point Cloud Neural Networks
    Zoumpekas, Thanasis
    Salamo, Maria
    Puig, Anna
    MODELING DECISIONS FOR ARTIFICIAL INTELLIGENCE, MDAI 2022, 2022, 13408 : 156 - 167
  • [29] STRUCTURE-AWARE GRAPH CONSTRUCTION FOR POINT CLOUD SEGMENTATION WITH GRAPH CONVOLUTIONAL NETWORKS
    Wang, Shanghong
    Dai, Wenrui
    Xu, Mingxing
    Li, Chenglin
    Zou, Junni
    Xiong, Hongkai
    2020 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO (ICME), 2020,
  • [30] Graph pooling in graph neural networks: methods and their applications in omics studies
    Wang, Yan
    Hou, Wenju
    Sheng, Nan
    Zhao, Ziqi
    Liu, Jialin
    Huang, Lan
    Wang, Juexin
    ARTIFICIAL INTELLIGENCE REVIEW, 2024, 57 (11)