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. 60, Issue 4, 0410002(2023)

DECANet: Image Semantic Segmentation Method Based on Improved DeepLabv3+

Lu Tang1,1,2,2、">">, Liang Wan1,1,2、">*, Tingting Wang1,1,2,2、">">, and Shusheng Li1,1,2,2、">">
Author Affiliations
  • 1College of Computer Science and Technology, Guizhou University, Guiyang 550025, Guizhou, China
  • 2Institute of Computer Software and Theory, Guizhou University, Guiyang 550025, Guizhou, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (14)
    Equations (9)
    References (33)
    Cited By (2)
    Tools
    Paper Information
    Topics
    Figures & Tables(14)
    Structure of the DeepLabv3+ model

    Fig. 1. Structure of the DeepLabv3+ model

    Download full size
    Structure of the DECANet model

    Fig. 2. Structure of the DECANet model

    Download full size
    Standard convolution

    Fig. 3. Standard convolution

    Download full size
    Depthwise convolution

    Fig. 4. Depthwise convolution

    Download full size
    Pointwise convolution

    Fig. 5. Pointwise convolution

    Download full size
    Loss function graph for implementing semantic segmentation under different algorithms

    Fig. 6. Loss function graph for implementing semantic segmentation under different algorithms

    Download full size
    Visualization results on PASCAL VOC2012 validation set

    Fig. 7. Visualization results on PASCAL VOC2012 validation set

    Download full size
    Visualization results on Cityscapes validation set

    Fig. 8. Visualization results on Cityscapes validation set

    Download full size

    • Table 1. Machine software and hardware configurations

      View table

      Table 1. Machine software and hardware configurations

      ProjectDetail
      CPUAMD EPYC 7742 64-Core Processor
      RAM32G
      GPUNVIDIA Tesla A100 40G
      Operating systemUbuntu 18.04.1
      CUDACuda 11.0
      Data processingPython 3.6

    • Table 2. Comparison results on PASCAL VOC2012 validation set

      View table

      Table 2. Comparison results on PASCAL VOC2012 validation set

      BackbonePSECANetMIoU /%
      ResNet-101170.31
      √72.73
      SENet-1011，6，12，1876.77
      √78.96
      ResNet-1011，6，12，1878.85
      √81.08

    • Table 3. IoU of different models

      View table

      Table 3. IoU of different models

      CategoryDECANet_1DECANet_2DECANet_3DECANet
      MIoU79.4879.7380.8081.08
      background94.6694.8595.1395.10
      bicycle44.2444.4644.1544.74
      boat77.4571.1277.2073.92
      bus95.5195.2095.8196.03
      cat92.0794.5494.1394.07
      cow90.3488.6692.2092.31
      dog89.5091.4590.4190.09
      motorbike86.6087.6585.9387.21
      pottedplant67.0968.0468.0367.05
      sofa50.6148.6559.5755.46
      tvmonitor75.7579.3175.5479.93
      aeroplane92.4892.9292.8193.09
      bird91.6591.2189.5990.97
      bottle80.2482.2681.9781.19
      car88.8689.8488.7089.78
      chair42.6744.0947.6551.09
      diningtable56.7953.9155.3957.08
      horse87.1087.8990.1991.70
      person88.0588.5688.2488.71
      sheep91.3789.7491.4591.91
      train86.0790.0592.7591.34

    • Table 4. Comparison results of MIoU values of different algorithms on PASCAL VOC2012 validation set

      View table

      Table 4. Comparison results of MIoU values of different algorithms on PASCAL VOC2012 validation set

      MethodBackbone networkMIoU /%
      FCN［5］VGG1662.20
      DeepLabv1［9］VGG1668.70
      DeepLabv2［10］ResNet10171.60
      DeconvNet［7］VGG1672.50
      DeepLabv3［12］ResNet10177.21
      DeepLabv3+［13］ResNet10178.85

      APCNet［15］

      WaveSNet［18］

      ResNet101

      ResNet101

      80.71

      79.90

      DECANet_1ResNet10179.48
      DECANet_2ResNet10179.73
      DECANet_3ResNet10180.80
      DECANetResNet10181.08

    • Table 5. IoU of two algorithms

      View table

      Table 5. IoU of two algorithms

      CategoryDeepLabv3+DECANet
      MIoU0.7420.760
      road0.9780.980
      sidewalk0.8260.842
      building0.9170.921
      wall0.4880.483
      fence0.5560.568
      pole0.5910.637
      traffic light0.6540.694
      traffic sign0.7460.780
      vegetation0.9200.921
      terrain0.6230.618
      sky0.9440.949
      person0.8020.817
      rider0.5940.633
      car0.9430.946
      truck0.7450.747
      bus0.7890.837
      train0.6370.666
      motorcycle0.6020.648
      bicycle0.7520.764

    • Table 6. Comparison results of MIoU value of different algorithms on Cityscapes validation set

      View table

      Table 6. Comparison results of MIoU value of different algorithms on Cityscapes validation set

      MethodMIoU /%
      FCN［5］63.1
      DeepLabv1［9］63.1
      BiSenet［14］69.0
      DFANet［16］71.3

      DeepLabv3+［13］

      N-Deeplabv3+［19］

      73.9

      74.6

      DECANet76.0
    Tools

    Get Citation

    Copy Citation Text

    Lu Tang, Liang Wan, Tingting Wang, Shusheng Li. DECANet: Image Semantic Segmentation Method Based on Improved DeepLabv3+[J]. Laser & Optoelectronics Progress, 2023, 60(4): 0410002

    Download Citation

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

    Category: Image Processing

    Received: Oct. 11, 2021

    Accepted: Dec. 21, 2021

    Published Online: Feb. 14, 2023

    The Author Email: Wan Liang (wanliangtr@163.com)

    DOI:10.3788/LOP212704

    Topics

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