Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 58, Issue 16, 1610008(2021)

Image Semantic Segmentation Method Based on Improved DeepLabv3+ Network

Cong Xu* and Li Wang
Author Affiliations
  • School of Computer and Information Engineering, Tianjin Chengjian University, Tianjin 300384, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (11)
    Equations (0)
    References (20)
    Cited By (7)
    Tools
    Paper Information
    Topics
    Figures & Tables(11)
    Structure of the DeepLabv3+ network

    Fig. 1. Structure of the DeepLabv3+ network

    Download full size
    Structure of the improved DeepLabv3+ network

    Fig. 2. Structure of the improved DeepLabv3+ network

    Download full size
    Structure of the ASPP module

    Fig. 3. Structure of the ASPP module

    Download full size
    Flow chart of the depth separable convolution

    Fig. 4. Flow chart of the depth separable convolution

    Download full size
    Structure of the FPN

    Fig. 5. Structure of the FPN

    Download full size
    Segmentation results of different methods. (a) Input image; (b) labeled image; (c) DeepLabv3+ network; (d) Ours 1

    Fig. 6. Segmentation results of different methods. (a) Input image; (b) labeled image; (c) DeepLabv3+ network; (d) Ours 1

    Download full size

    • Table 1. Hardware and software configurations of the machine

      View table

      Table 1. Hardware and software configurations of the machine

      ProjectDetail
      CPUInter i7-4770k
      RAM32 G
      Graphics cardTITAN V
      Operating system64-bit Ubuntu 18.04.3
      CUDACuda 10.2
      Data processingPytorch 3.6

    • Table 2. Test results under different combinations of dilation rate in ASPP module unit: %

      View table

      Table 2. Test results under different combinations of dilation rate in ASPP module unit: %

      Dilation ratemIoU
      (6,12,18)79.73
      (6,12,18,24)79.80
      (4,8,12,16)79.97

    • Table 3. Training time and mIoU of different networks

      View table

      Table 3. Training time and mIoU of different networks

      Convolution method in ASPPTime/hmIoU/%
      Standard convolution23.279.97
      Depthwise separable convolution17.379.46

    • Table 4. Performances of different networks

      View table

      Table 4. Performances of different networks

      NetworkParameter /MFLOPs /GMACSpeed /(frame·s-1)
      DeepLabv261.4175.4012.55
      DeepLabv358.0471.1613.07
      DeepLabv3+59.3492.9313.45
      Ours 164.6599.5312.66
      Ours 247.9781.3416.96

    • Table 5. mIoU of different networks on the PASCAL VOC2012 verification set

      View table

      Table 5. mIoU of different networks on the PASCAL VOC2012 verification set

      NetworkmIoU/%
      DeepLabv168.70
      DeepLabv276.35
      DeepLabv377.21
      DeepLabv3+78.85
      Ours 279.46
      Ours 179.97
    Tools

    Get Citation

    Copy Citation Text

    Cong Xu, Li Wang. Image Semantic Segmentation Method Based on Improved DeepLabv3+ Network[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1610008

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Set citation alerts for article
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Oct. 24, 2020

    Accepted: Dec. 8, 2020

    Published Online: Aug. 19, 2021

    The Author Email: Cong Xu (769458796@qq.com)

    DOI:10.3788/LOP202158.1610008

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology