Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 1, 48(2024)

Camouflaged object segmentation based on edge enhancement and feature fusion

Mingyan LI1,2, Chuan WU1,2、*, and Ming ZHU1,2
Author Affiliations
  • 1Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033,China
  • 2University of Chinese Academy of Sciences,Beijing 100049,China
  • show less
    Figures & Tables(11)
    Network structure diagram
    Multi-scale feature enhanced module
    Cross-level fusion module
    Cross-layer attention module
    Vision comparison of our method with other methods
    P-R curves and F-measure curves of 10 different methods on four benchmark datasets. Our method is shown with a solid red line. The closer the P-R curve is to the upper right corner and the higher the F-measure curve is,the better the performance of the model is.
    Vision comparison of removed different modules
    • Table 1. Quantitative results of different models for four evaluation metrics on four dataset(CHAMELEON,CAMO-test,COD10k-test,NC4K)

      View table
      View in Article

      Table 1. Quantitative results of different models for four evaluation metrics on four dataset(CHAMELEON,CAMO-test,COD10k-test,NC4K)

      基线模型CHAMELEONCAMO-test
      SαEØFβwMAE↓SαEØFβwMAE↓
      BASNet0.6870.7210.4740.1180.6180.6610.4130.159
      EGNet0.8480.8700.7020.0500.7320.7680.5830.104
      CPD0.8530.8660.7060.0520.7260.7290.5500.115
      F3Net0.8480.9170.7440.0470.7110.7800.5640.109
      PraNet0.8600.9070.7600.0440.7690.8240.6630.094
      SINet0.8690.8910.7400.0440.7510.7710.6060.100
      PFNet0.8820.9420.8100.0330.7820.8520.6950.085
      C2FNet0.8880.9460.8280.0320.8200.8640.7520.066
      SINetV20.8880.9420.8160.0300.8200.8820.7430.070
      LSR0.8900.9480.8200.0300.7870.8520.6960.080
      UGTR0.8880.9180.7940.0310.7850.8590.6860.086
      YOLOv5m-seg0.8340.8840.7670.0450.6990.7110.5560.098
      YOLOv5x-seg0.8510.9020.7960.0420.7400.7890.5920.087
      本文方法0.8920.9520.8280.0270.8280.8890.8170.061
      基线模型COD10K-testNC4K
      SαEØFβwMAE↓SαEØFβwMAE↓
      BASNet0.6340.6780.3650.1050.6950.7850.5460.095
      EGNet0.7370.7790.5090.0560.7770.8640.6390.075
      CPD0.7470.7700.5080.0590.7870.8520.6960.072
      F3Net0.7390.8190.5440.0510.7800.8480.6560.070
      PraNet0.7890.8610.6290.0450.8220.8770.7240.059
      SINet0.7710.8060.5510.0510.8080.8760.7230.058
      PFNet0.8000.8680.6600.0400.8390.8920.7450.053
      C2FNet0.8130.8900.6360.0360.8380.8980.7620.049
      SINetV20.8150.8870.6800.0370.8400.9070.7700.048
      LSR0.8040.8890.6730.0370.8400.9070.7700.048
      UGTR0.8180.8500.6670.0350.8390.8920.7460.052
      YOLOv5m-seg0.7150.7320.5270.0480.7530.7800.6340.066
      YOLOv5x-seg0.7940.8320.5900.0440.8020.8680.7150.053
      本文方法0.8220.8900.6730.0340.8500.9080.7750.044
    • Table 2. Speed and model complexity analysis on multiple models

      View table
      View in Article

      Table 2. Speed and model complexity analysis on multiple models

      Method本文方法SINetPFNetUGTRMGL-RLSRSINetV2
      FLOPs/G21.2638.0419.001 024549.6266.6431.26
      Params./M29.4748.9549.5048.8763.5850.9524.93
      FPS44.2566218125625
    • Table 3. Effectiveness analysis of different modules

      View table
      View in Article

      Table 3. Effectiveness analysis of different modules

      边缘提取模块多尺度特征增强模块跨层级特征聚合模块注意力模块COD10K-test
      SαEØFβwMAE↓
      0.7660.8280.5520.045
      0.8030.8570.6390.040
      0.8120.8640.6600.038
      0.8120.8680.6520.038
      0.8190.8750.6640.036
      0.8160.8720.6660.036
      0.8220.8900.6730.034
    • Table 4. Effect of the proportion of the two loss functions on network performance

      View table
      View in Article

      Table 4. Effect of the proportion of the two loss functions on network performance

      λ1λ2COD10K-test
      SαEØFβwMAE↓
      110.8130.8870.6600.036
      210.8200.8900.6690.034
      510.8220.8900.6730.034
      1010.8250.8840.6700.035
      120.8070.8780.6620.035
      150.8000.8710.6540.037
    Tools

    Get Citation

    Copy Citation Text

    Mingyan LI, Chuan WU, Ming ZHU. Camouflaged object segmentation based on edge enhancement and feature fusion[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(1): 48

    Download Citation

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

    Category: Research Articles

    Received: Feb. 20, 2023

    Accepted: --

    Published Online: Mar. 27, 2024

    The Author Email: Chuan WU (wuchuan0458@sina.com)

    DOI:10.37188/CJLCD.2023-0064

    Topics