Optics and Precision Engineering, Volume. 32, Issue 6, 843(2024)
Image super-resolution network based on multi-scale adaptive attention
Figures & Tables(16)
Fig. 8. Comparison of visual effect of each method with a scale factor of ×4
Table 1. Ablation experiments of the numbers of MFFB
View tableView in ArticleTable 1. Ablation experiments of the numbers of MFFB
数量 6 7 8 9 参数量/M 7.36 8.52 9.61 10.83 PSNR 38.01 38.20 38.37 38.40
Table 2. Results of ablation experiments of the structure of BU
View tableView in ArticleTable 2. Results of ablation experiments of the structure of BU
结构 Set5 Set14 BSD100 Urban100 PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM 结构1 32.35 0.896 3 28.66 0.782 1 27.45 0.727 4 26.58 0.799 8 结构2 32.39 0.895 9 28.78 0.787 6 27.42 0.729 4 26.64 0.803 3 结构3 32.43 0.899 1 28.85 0.798 2 27.54 0.731 9 26.59 0.802 8 结构4 32.69 0.900 2 29.03 0.810 5 27.74 0.745 1 26.92 0.816 7 结构5 32.76 0.901 8 29.11 0.811 8 27.83 0.746 3 27.03 0.819 5 结构6 32.77 0.902 0 29.12 0.812 0 27.85 0.746 9 27.08 0.819 4 完整的BU 32.79 0.902 3 29.15 0.812 8 27.87 0.747 3 27.12 0.820 9
Table 3. Results of ablation experiments of the structure of ADB
View tableView in ArticleTable 3. Results of ablation experiments of the structure of ADB
分支 3×3 5×5 7×7 3×3+5×5 3×3+5×5+7×7 3×3+5×5+AWU 参数量/M 3.21 5.84 8.53 9.50 19.99 9.69 PSNR 34.26 34.37 34.43 34.81 34.95 34.91
Table 4. Ablation experiments of the groups of MPIB
View tableView in ArticleTable 4. Ablation experiments of the groups of MPIB
组数 2 4 8 参数量/M 7.79 9.61 14.55 PSNR 37.82 38.37 38.53
Table 5. Comparison of the evaluation metrics of different attention
View tableView in ArticleTable 5. Comparison of the evaluation metrics of different attention
方法 参数量/M Set5 Set14 BSD100 Urban100 PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM 基础网络 9.65 34.66 0.928 8 30.52 0.847 4 29.12 0.809 7 28.91 0.872 3 +ECA 9.67 34.72 0.929 8 30.67 0.848 3 29.21 0.811 0 29.01 0.872 4 +CBAM 9.73 34.78 0.930 1 30.70 0.848 7 29.26 0.811 4 29.09 0.873 1 +BAM 9.78 34.76 0.929 9 30.72 0.849 3 29.22 0.811 5 29.18 0.873 6 +ADA 9.69 34.93 0.931 7 30.80 0.850 9 29.40 0.813 6 29.37 0.875 3
Table 6. Comparison of evaluation metrics of each method with scale factors of ×2， ×3， ×4
View tableView in ArticleTable 6. Comparison of evaluation metrics of each method with scale factors of ×2， ×3， ×4
比例 方法 参数量/M Set5 Set14 BSD100 Urban100 PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM ×3 EDSR 43.7 34.65 0.928 0 30.52 0.846 2 29.25 0.809 3 28.80 0.865 3 RDN 22.3 34.71 0.929 6 30.57 0.846 8 29.26 0.809 3 28.80 0.865 3 E-GSCN 15.9 34.69 0.929 4 30.59 0.847 1 29.28 0.809 9 28.87 0.866 9 DRLN 33.7 34.78 0.930 3 30.73 0.848 8 29.36 0.811 7 29.21 0.872 2 SwinIR 11.9 34.89 0.931 2 30.77 0.850 3 29.37 0.812 4 29.29 0.874 4 MAAN 9.69 34.93 0.931 7 30.80 0.850 9 29.40 0.813 6 29.37 0.875 3 Bicubic - 28.43 0.802 2 26.10 0.693 6 25.97 0.651 7 23.14 0.659 9 SRCNN 0.02 30.48 0.862 8 27.49 0.750 3 26.90 0.710 1 24.52 0.722 1 VDSR 0.67 31.35 0.883 8 28.01 0.767 4 27.29 0.725 1 25.18 0.752 4 DRRN 0.30 31.68 0.888 8 28.21 0.772 0 27.38 0.728 4 25.44 0.763 8 A2F-L 1.37 32.32 0.896 4 28.67 0.783 9 27.62 0.737 9 36.32 0.793 1 ×4 JTF-SISR 18.5 32.53 0.898 9 28.80 0.786 7 27.70 0.741 0 26.62 0.802 2 EDSR 43.1 32.46 0.896 8 28.80 0.787 6 27.71 0.742 0 26.64 0.803 3 RDN 22.6 32.47 0.899 0 28.81 0.787 1 27.72 0.741 9 26.61 0.802 8 E-GSCN 15.9 32.53 0.899 2 28.84 0.787 9 27.74 0.742 4 26.71 0.805 0 DRLN 34.0 32.63 0.900 2 28.94 0.790 0 27.83 0.744 4 26.98 0.811 9 SwinIR 11.9 32.75 0.902 1 28.94 0.791 4 27.83 0.745 9 27.07 0.816 4 MAAN 9.72 32.79 0.902 3 29.15 0.812 8 27.87 0.747 3 27.12 0.820 9
Table 6. Comparison of evaluation metrics of each method with scale factors of ×2， ×3， ×4
View tableView in ArticleTable 6. Comparison of evaluation metrics of each method with scale factors of ×2， ×3， ×4
比例 方法 参数量/M Set5 Set14 BSD100 Urban100 PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM Bicubic - 33.69 0.928 4 30.34 0.867 5 29.57 0.843 4 26.88 0.843 8 SRCNN 0.02 36.66 0.954 2 32.42 0.906 3 31.36 0.887 9 29.50 0.894 6 VDSR 0.67 37.53 0.958 7 33.03 0.912 4 31.90 0.896 0 30.76 0.914 0 DRRN 0.30 37.74 0.959 1 33.23 0.913 6 32.05 0.897 3 31.23 0.918 8 A2F-L 1.36 38.09 0.960 7 33.78 0.919 2 32.23 0.900 2 32.46 0.931 3 JTF-SISR 17.8 38.24 0.961 3 33.83 0.919 6 32.34 0.901 7 32.81 0.934 9 ×2 EDSR 40.7 38.11 0.960 2 33.92 0.919 5 32.32 0.901 3 32.93 0.935 1 RDN 22.1 38.24 0.961 4 34.01 0.921 2 32.34 0.901 7 32.89 0.935 3 E-GSCN 15.7 38.21 0.961 2 33.95 0.920 3 32.35 0.901 9 32.94 0.935 7 DRLN 32.4 38.27 0.961 6 34.28 0.923 1 32.44 0.902 8 33.37 0.939 0 SwinIR 11.8 38.35 0.962 0 34.14 0.922 7 32.44 0.903 0 33.40 0.939 3 MAAN 9.61 38.37 0.962 6 34.26 0.923 8 32.49 0.903 2 33.43 0.940 1 Bicubic - 30.41 0.865 5 27.64 0.772 2 27.21 0.734 4 24.46 0.741 1 SRCNN 0.02 32.75 0.909 0 29.28 0.820 9 28.41 0.786 3 26.24 0.798 9 VDSR 0.67 33.66 0.921 3 29.77 0.831 4 28.82 0.797 6 27.14 0.827 9 DRRN 0.30 34.03 0.924 4 29.96 0.834 9 28.95 0.800 4 27.53 0.837 8 A2F-L 1.37 34.54 0.928 3 30.41 0.843 6 29.14 0.806 2 28.40 0.857 4 JTF-SISR 18.1 34.72 0.929 5 30.57 0.846 5 29.25 0.809 0 28.74 0.864 5
Table 7. Complexity comparison of the 3 methods
View tableView in ArticleTable 7. Complexity comparison of the 3 methods
方法 PSNR SSIM 时间
/s
参数量/M 计算量/G EDSR 27.71 0.742 0 0.338 43.1 516 RDN 27.72 0.741 9 0.224 22.6 209 E-GSCN 27.74 0.742 4 0.136 15.9 144 DRLN 27.83 0.744 4 0.271 34.0 292 SwinIR 27.83 0.745 9 0.112 11.9 128 MAAN 27.87 0.747 3 0.085 9.72 79
Tools
Get Citation
Copy Citation Text
Ying ZHOU, Shenghu PEI, Haiyong CHEN, Shibo XU. Image super-resolution network based on multi-scale adaptive attention[J]. Optics and Precision Engineering, 2024, 32(6): 843
Paper Information
Category:
Received: Jul. 23, 2023
Accepted: --
Published Online: Apr. 19, 2024
The Author Email: ZHOU Ying (zhouying2007@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20