Figures (8)  Tables (4)

    • Figure 1. 

      The LLIE framework.

    • Figure 2. 

      FA-UNet.

    • Figure 3. 

      Depthwise separable convolution network.

    • Figure 4. 

      Convolutional block attention module.

    • Figure 5. 

      Visual comparison between dark light and enhanced image.

    • Figure 6. 

      Visual comparison on real-world image.

    • Figure 7. 

      Visual comparison of different loss functions

    • Figure 8. 

      Visual effects of LLIE methods.

    • Methods PSNR SSIM NIQE LPIPS
      LOL-v1 RetinexNet 16.77 0.43 8.734 0.381
      DCE-net 15.14 0.70 7.755 0.330
      EnlightenGAN 18.63 0.812 6.869 0.412
      SCI 14.77 0.679 6.646 0.335
      FA-UNet 17.22 0.71 5.879 0.318
      SICE RetinexNet 15.84 0.73 4.368 0.407
      DCE-net 16.12 0.87 3.987 0.362
      EnlightenGAN 13.94 0.605 2.683 0.253
      SCI 13.15 0.523 3.360 0.317
      FA-UNet 19.88 0.89 3.175 0.308

      Table 1. 

      Image quality evaluation table.

    • Loss function combination PSNR SSIM
      $ w/o{L}_{spa} $ 17.04 0.68
      $ w/o $ $ {L}_{ill} $ 15.79 0.59
      $ w/o $ $ {L}_{exp} $ 16.31 0.65
      $ {L}_{Total} $ 17.22 0.71

      Table 2. 

      PSNR/SSIM of different combinations of loss functions

    • Image sequence Length Low-light RetinexNet HE DCE-net FA-UNet
      MH_04_difficult 91.747 Fail 0.216 0.114 0.125 0.070
      MH_05_difficult 97.593 Fail 0.198 0.829 0.100 0.042
      V1_03_difficult 78.982 Fail 0.098 0.100 0.122 0.087
      Syn_KITTI_Seq00 3,724.187 16.012 4.403 5.135 4.809 3.009
      Syn_KITTI_Seq02 5,067.223 Fail 16.109 8.885 9.136 6.891
      Syn_KITTI_Seq03 560.888 0.668 0.462 0.272 0.357 0.195
      Syn_KITTI_Seq09 1,705.051 Fail 5.207 4.902 3.667 3.017

      Table 3. 

      Comparison results of the root mean square error of the absolute trajectory error (m).

    • Methods RetinexNet DCE-net FA-UNet
      KITTI Euroc KITTI Euroc KITTI Euroc
      FLOPs (109) 13.721 10.623 0.059 0.046 0.034 0.022
      Test time (ms) 72.42 56.05 35.24 30.54 15.86 12.72

      Table 4. 

      Model efficiency evaluation indices