Acta Photonica Sinica, Volume. 53, Issue 8, 0810002(2024)

Camouflage Object Detection Based on Feature Fusion and Edge Detection

Cheng DING... Xueqiong BAI*, Yong LV*, Yang LIU, Chunhui NIU and Xin LIU |Show fewer author(s)
Author Affiliations
  • School of Instrumentation Science and Opto-electronics Engineering, Beijing Information Science and Technology University, Beijing 100192, China
  • show less
    Figures & Tables(12)
    Overall architecture of F2-EDNet
    Feature enhance module
    MFAM module structure
    PR curves of different models on CAMO, COD10K, NC4K datasets
    Qualitative results of different models on test datasets
    • Table 1. Datasets settings

      View table
      View in Article

      Table 1. Datasets settings

      DatasetsImagesTrainingTesting
      CAMO111 2501 000250
      COD10K125 0663 0402 026
      NC4K134 12104 121
    • Table 2. Comparison of performance metrics of different models on three datasets

      View table
      View in Article

      Table 2. Comparison of performance metrics of different models on three datasets

      ModelNC4K
      SαFβwMAE↓Eφ
      SINet (2020)0.8080.7230.0580.883
      PFNet (2021)0.8290.7450.0530.894
      C2FNet (2021)0.8380.7620.0490.901
      SINetV2 (2022)0.8470.7700.0480.901
      SegMaR (2022)0.8410.7810.0460.905
      OCENet (2022)0.8530.7850.0450.908
      BGNet (2022)0.8510.7880.0440.911
      ZoomNet (2022)0.8530.7840.0430.907
      FEDER (2023)0.8470.7890.0440.913
      DGNet (2023)0.8570.7840.0420.910
      F2-EDNet0.8660.7930.0410.915
    • Table 3. Comparative analysis of the efficiency of model runs

      View table
      View in Article

      Table 3. Comparative analysis of the efficiency of model runs

      TypeModelTotal mult-adds(G)↓Params size (MB)↓Average speed/(frame·s-1)↑
      Single taskSINet38.60195.7935.266
      PFNet37.79185.9943.416
      C²FNet26.02105.4534.771
      SINetV224.3499.7138.058
      ZoomNet101.91130.2421.861
      Multi taskSegMaR66.98222.4848.828
      OCENet50.20219.9441.045
      BGNet83.40311.2139.865
      FEDER71.60168.3716.203
      DGNet5.347.2040.000
      F2-EDNet38.87442.1946.151
    • Table 4. Results of modular ablation experiments

      View table
      View in Article

      Table 4. Results of modular ablation experiments

      ExperimentFEMCGEBMFAMDSNC4K
      SαFβwMAE↓Eφ
      10.8510.7540.0500.904
      20.8570.7670.0470.906
      30.8520.7700.0460.907
      40.8620.7750.0450.909
      50.8640.7900.0420.912
      60.8660.7930.0410.915
    • Table 5. Module efficiency ablation experiment

      View table
      View in Article

      Table 5. Module efficiency ablation experiment

      NoFEMCGEBMFAMDSTotal mult-adds(G)Params size (MB)Average speed/(frame·s-1
      13.180350.68087.111
      211.970358.77057.238
      337.300440.94055.228
      438.860442.19047.880
      538.870442.19046.151
    • Table 6. Ablation experiments on the number of feature enhancement module

      View table
      View in Article

      Table 6. Ablation experiments on the number of feature enhancement module

      f0f1f2f3NC4K
      SαFβwMAE↓Eφ
      0.8530.7560.0490.905
      0.8570.7620.0480.904
      0.8570.7680.0460.903
      0.8570.7670.0470.906
    • Table 7. Deeply supervised quantity ablation experiments

      View table
      View in Article

      Table 7. Deeply supervised quantity ablation experiments

      f1f2f3NC4K
      SαFβwMAE↓Eφ
      0.7900.0420.9120.790
      0.7810.0440.9080.781
      0.7930.0410.9150.793
      0.7820.0450.9130.782
    Tools

    Get Citation

    Copy Citation Text

    Cheng DING, Xueqiong BAI, Yong LV, Yang LIU, Chunhui NIU, Xin LIU. Camouflage Object Detection Based on Feature Fusion and Edge Detection[J]. Acta Photonica Sinica, 2024, 53(8): 0810002

    Download Citation

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

    Category:

    Received: Jan. 18, 2024

    Accepted: Mar. 27, 2024

    Published Online: Oct. 15, 2024

    The Author Email: BAI Xueqiong (bxq@bistu.edu.cn), LV Yong (lvyong@bistu.edu.cn)

    DOI:10.3788/gzxb20245308.0810002

    Topics