Optics and Precision Engineering, Volume. 32, Issue 8, 1212(2024)

Real-time urban street view semantic segmentation based on cross-layer aggregation network

Zhiqiang HOU1...2, Minjie CHENG1,2,*, Sugang MA1,2, Minjie QU1,2, and Xiaobao YANG12 |Show fewer author(s)
Author Affiliations
  • 1Xi'an University of Posts and Telecommunications, Institute of Computer, Xi'an702, China
  • 2Xi'an University of Posts and Telecommunications, Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing, Xi'an71011, China
  • show less
    Figures & Tables(12)
    Overall structure of CLANet (Cross-Layer Aggregation Network, CLANet)
    Cross-Layer Aggregation Module
    Comparison diagram between DAPPM and CLA-PPM
    Multi-Scale Fusion Module, MSFM
    Accuracy-speed comparison on the Cityscapes test set
    Accuracy-speed comparison on the CamVid test set
    Visual segmentation results of the Cityscapes dataset
    Visual segmentation results of the CamVid dataset
    • Table 1. Ablation study of CLA-PPM on the Cityscapes validation

      View table
      View in Article

      Table 1. Ablation study of CLA-PPM on the Cityscapes validation

      BaselineDAPPMCLA-PPMFLOPs/GParams/MmIoU/%Speed/FPS
      Sparsity=2Sparsity=3
      97.714.274.596
      98.815.5375.283
      98.014.6875.089
      97.914.5274.990
      98.215.075.385
    • Table 2. Ablation study of CLA-Net on the Cityscapes validation

      View table
      View in Article

      Table 2. Ablation study of CLA-Net on the Cityscapes validation

      BaselineCLAMCLA-PPMMSFMFLOPs/GParams/MmIoU/%Speed /FPS
      97.714.274.596
      103.114.575.286
      98.215.075.385
      98.314.2274.893
      103.615.3275.777
      98.815.0475.583
      103.714.5375.484
      104.315.3576.075
    • Table 3. Comparison of accuracy and speed of Cityscapes

      View table
      View in Article

      Table 3. Comparison of accuracy and speed of Cityscapes

      MethodReferenceResolutionmIoU/%

      #FPS

      (PyTorch)

      #FPS

      (TensorRT)

      ValTest
      GAS19CVPR2020769×1 53771.8108.4
      HMSeg20BMVC2020768×1 53674.383.2
      DCNet21ICPR2021512×1 02471.2142
      HyperSeg-M22CVPR20211 024×2 04876.275.836.9
      RELAXNet23Neurocomputing2022512×1 02474.864
      FPANet C24APPL INTELL20221 024×2 04875.931
      BiAttnNet25SPL2022512 × 1 02474.789.2
      LETNet26T-ITS2023512×1 02472.8150
      SRDENet27IET COMPUT VIS2023512×1 02475.465
      BiSeNetV228IJCV2021512×1 02473.472.6156
      BiSeNetV2-L28IJCV2021512×1 02475.875.347.3
      FasterSeg29arXiv20191 024×2 04873.171.5163.9
      STDC1-Seg509CVPR2021512×1 02472.271.9250.4
      STDC1-Seg759CVPR2021768×1 53674.575.3126.7
      CPANet-T5030CAC 2022512 × 1 02472.5234.5
      BiSeNetV3-5031Neurocomputing2023512×1 02473.473.5244.3
      CLANet-50(Ours)512×1 02473.373.0143294
      CLANet-75(Ours)768×1 53676.075.875164
    • Table 4. Comparison of accuracy and speed of CamVid

      View table
      View in Article

      Table 4. Comparison of accuracy and speed of CamVid

      MethodReferenceResolutionmIoU/%

      #FPS

      (PyTorch)

      #FPS

      (TensorRT)

      CAS32CVPR2019720×96071.8169
      GAS19CVPR2020720×96072.8153.1
      DSANet33ExpertSyst. Appl. 2021720×96069.975.3
      FSFNet34IEEE T INSTRUM MEAS2021720×96075.191
      RELAXNet23Neurocomputing2022720×96071.279
      FPANet B24APPL INTELL2022720×96072.988
      LETNet26T-ITS2023720×96070.5200
      SRDENet27IET COMPUT VIS2023720×96074.878.3
      BiSeNetV228IJCV2021720×96072.4124.5
      BiSeNetV2-L28IJCV2021720×96073.232.7
      STDC1-Seg509CVPR2021720×96073.0197.6
      CPANet-T30CAC 2022720×96073.9213.9
      BiSeNetV331Neurocomputing2023720×96075.1198.4
      CLANet(Ours)720×96074.8116239
    Tools

    Get Citation

    Copy Citation Text

    Zhiqiang HOU, Minjie CHENG, Sugang MA, Minjie QU, Xiaobao YANG. Real-time urban street view semantic segmentation based on cross-layer aggregation network[J]. Optics and Precision Engineering, 2024, 32(8): 1212

    Download Citation

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

    Category:

    Received: Oct. 21, 2023

    Accepted: --

    Published Online: May. 29, 2024

    The Author Email: CHENG Minjie (rebu1999@163.com)

    DOI:10.37188/OPE.20243208.1212

    Topics