Lightweight Feature Point and Descriptor Extraction Network Based on SuperPoint
Author:
Affiliation:

Clc Number:

Fund Project:

  • Article
  • |
  • Figures
  • |
  • Metrics
  • |
  • Reference
  • |
  • Related
  • |
  • Cited by
  • |
  • Materials
  • |
  • Comments
    Abstract:

    The extraction of image feature points and descriptors is the foundation of some tasks such as SLAM, SFM, and 3D reconstruction. Preeminent algorithms for image feature point and descriptor extraction play a significant role in processing these tasks. This study accomplishes some improvements in the SuperPoint network with high robustness and good performance in the extraction of feature points and descriptors. Considering the flaws of the heavy calculation burden and massive parameters, the authors first change the ordinary convolution to depthwise separable convolution, the number of layers, and the down-sampling method. Afterward, the channel pruning algorithm is perfected so that it can be applied to depthwise separable convolution and prune the network. Experiments have proved that this study reduces the network parameter number and calculation burden respectively to 15% and 5% those of the original SuperPoint network, and the FPS is increased by 6.64 times under the condition of a slight loss of feature point detection and matching effects. Thus, good experimental results are achieved.

    Reference
    Related
    Cited by
Get Citation

李志强,朱明.基于SuperPoint的轻量级特征点及描述子提取网络.计算机系统应用,2021,30(11):310-316

Copy
Share
Article Metrics
  • Abstract:
  • PDF:
  • HTML:
  • Cited by:
History
  • Received:January 30,2021
  • Revised:February 26,2021
  • Adopted:
  • Online: October 22,2021
  • Published:
You are the firstVisitors
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-3
Address:4# South Fourth Street, Zhongguancun,Haidian, Beijing,Postal Code:100190
Phone:010-62661041 Fax: Email:csa (a) iscas.ac.cn
Technical Support:Beijing Qinyun Technology Development Co., Ltd.

Beijing Public Network Security No. 11040202500063