Acta Photonica Sinica, Volume. 53, Issue 8, 0810002(2024)
Camouflage Object Detection Based on Feature Fusion and Edge Detection
Figures & Tables(12)
Fig. 4. PR curves of different models on CAMO， COD10K， NC4K datasets
Table 1. Datasets settings
View tableView in ArticleTable 1. Datasets settings
Datasets Images Training Testing CAMO［ 11 ］1 250 1 000 250 COD10K［ 12 ］5 066 3 040 2 026 NC4K［ 13 ］4 121 0 4 121
Table 2. Comparison of performance metrics of different models on three datasets
View tableView in ArticleTable 2. Comparison of performance metrics of different models on three datasets
Model NC4K Sα↑ Fβw↑ MAE↓ Eφ↑ SINet （2020） 0.808 0.723 0.058 0.883 PFNet （2021） 0.829 0.745 0.053 0.894 C2FNet （2021） 0.838 0.762 0.049 0.901 SINetV2 （2022） 0.847 0.770 0.048 0.901 SegMaR （2022） 0.841 0.781 0.046 0.905 OCENet （2022） 0.853 0.785 0.045 0.908 BGNet （2022） 0.851 0.788 0.044 0.911 ZoomNet （2022） 0.853 0.784 0.043 0.907 FEDER （2023） 0.847 0.789 0.044 0.913 DGNet （2023） 0.857 0.784 0.042 0.910 F2-EDNet 0.866 0.793 0.041 0.915
Table 3. Comparative analysis of the efficiency of model runs
View tableView in ArticleTable 3. Comparative analysis of the efficiency of model runs
Type Model Total mult-adds（G）↓ Params size （MB）↓ Average speed/（frame·s-1）↑ Single task SINet 38.60 195.79 35.266 PFNet 37.79 185.99 43.416 C²FNet 26.02 105.45 34.771 SINetV2 24.34 99.71 38.058 ZoomNet 101.91 130.24 21.861 Multi task SegMaR 66.98 222.48 48.828 OCENet 50.20 219.94 41.045 BGNet 83.40 311.21 39.865 FEDER 71.60 168.37 16.203 DGNet 5.34 7.20 40.000 F2-EDNet 38.87 442.19 46.151
Table 4. Results of modular ablation experiments
View tableView in ArticleTable 4. Results of modular ablation experiments
Experiment FEM CGEB MFAM DS NC4K Sα↑ Fβw↑ MAE↓ Eφ↑ 1 0.851 0.754 0.050 0.904 2 √ 0.857 0.767 0.047 0.906 3 √ √ 0.852 0.770 0.046 0.907 4 √ √ 0.862 0.775 0.045 0.909 5 √ √ √ 0.864 0.790 0.042 0.912 6 √ √ √ √ 0.866 0.793 0.041 0.915
Table 5. Module efficiency ablation experiment
View tableView in ArticleTable 5. Module efficiency ablation experiment
No FEM CGEB MFAM DS Total mult-adds（G） Params size （MB） Average speed/（frame·s-1） 1 3.180 350.680 87.111 2 √ 11.970 358.770 57.238 3 √ √ 37.300 440.940 55.228 4 √ √ √ 38.860 442.190 47.880 5 √ √ √ √ 38.870 442.190 46.151
Table 6. Ablation experiments on the number of feature enhancement module
View tableView in ArticleTable 6. Ablation experiments on the number of feature enhancement module
f0 f1 f2 f3 NC4K Sα↑ Fβw↑ MAE↓ Eφ↑ √ 0.853 0.756 0.049 0.905 √ √ 0.857 0.762 0.048 0.904 √ √ √ 0.857 0.768 0.046 0.903 √ √ √ √ 0.857 0.767 0.047 0.906
Table 7. Deeply supervised quantity ablation experiments
View tableView in ArticleTable 7. Deeply supervised quantity ablation experiments
f1 f2 f3 NC4K Sα↑ Fβw↑ MAE↓ Eφ↑ 0.790 0.042 0.912 0.790 √ 0.781 0.044 0.908 0.781 √ √ 0.793 0.041 0.915 0.793 √ √ √ 0.782 0.045 0.913 0.782
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
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)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20