• Optics and Precision Engineering
  • Vol. 32, Issue 3, 445 (2024)
Jingjing ZHANG1,2,3,*, Xingzhuo DU1,2,3, Shuai ZHI4,5, and Guopeng DING4,5,*
Author Affiliations
  • 1School of Automation, China University of Geosciences (Wuhan), Wuhan430074, China
  • 2Hubei Provincial Key Laboratory of Advanced Control and Intelligent Automation for Complex Systems, Wuhan430074, China
  • 3Engineering Research Center of Earth Exploration Intelligent Technology, Ministry of Education, Wuhan40074, China
  • 4Innovation Academy for Microsatellites of Chinese Academy of Sciences, Shanghai201203, China
  • 5Shanghai Microsatellite Engineering Center, Shanghai201203, China
  • show less
    DOI: 10.37188/OPE.20243203.0445 Cite this Article
    Jingjing ZHANG, Xingzhuo DU, Shuai ZHI, Guopeng DING. Atrous convolution and Bilateral grid network[J]. Optics and Precision Engineering, 2024, 32(3): 445 Copy Citation Text show less
    Architecture of AB-Net
    Fig. 1. Architecture of AB-Net
    Schematic diagram of atrous convolution with different expansion rates
    Fig. 2. Schematic diagram of atrous convolution with different expansion rates
    Schematic diagram of feature extraction module structure
    Fig. 3. Schematic diagram of feature extraction module structure
    Schematic diagram of stacked hourglass module of 3D CNN
    Fig. 4. Schematic diagram of stacked hourglass module of 3D CNN
    Schematic diagram of Bilateral grid
    Fig. 5. Schematic diagram of Bilateral grid
    Result of different methods on SceneFlow dataset
    Fig. 6. Result of different methods on SceneFlow dataset
    Result of different methods on SceneFlaw dataset
    Fig. 7. Result of different methods on SceneFlaw dataset
    Test results on Middlebury 2014
    Fig. 8. Test results on Middlebury 2014
    Untrained results on Middlebury 2014 dataset
    Fig. 9. Untrained results on Middlebury 2014 dataset
    Experimental results of real scene
    Fig. 10. Experimental results of real scene
    网络名称残差层简化ASPP双边格网Res-CVEpe/pixelt/ms
    PSM-Net1/41.092 310
    AA-Net1/40.871 147
    AB-Net1/41.161 125
    1/41.011 856
    1/80.86951
    Table 1. Results of ablation experiment
    网络名称Epe/pixelNumber of parameters/108
    GC-Net42.513.50
    PSM-Net51.095.20
    CRL[16]1.3278.77
    AA-Net[17]0.87-
    AED-Net[18]0.89-
    AB-Net0.863.20
    Table 2. Result of different methods on SceneFlow dataset

    网络

    名称

    All Pixel/%Non-Occluded Pixels/%

    Number of

    parameters/108

    D1-bgD1-fgD1-allD1-bgD1-fgD1-all
    GC-Net42.216.162.872.025.582.613.50
    PSM-Net51.864.622.321.714.312.145.20
    CRL131.995.392.554.932.321.8978.77
    AA-Net142.483.592.672.323.122.45-
    AB-Net1.914.342.261.824.172.113.20
    Table 3. Test result on KITTI 2015 binocular stereo matched dataset
    网 络ONoc/%OAll/%ANoc/%AAll/%Number of parameters/108
    GC-Net41.772.060.60.73.50
    PSM-Net121.491.890.50.65.20
    AA-Net142.543.180.60.7-
    AED-Net153.404.110.70.8-
    AB-Net1.442.450.60.73.20
    Table 4. Test results on KITTI 2012 binocular stereo matched dataset
    网 络Bad2.0/%
    PSM-Net518.58
    GA-Net617.43
    AA-Net1315.94
    AB-Net7.56
    Table 5. Result on middlebury 2014 dataset analysis
    网 络Bad2.0/%
    PSM-Net524.8
    GA-Net619.1
    AA-Net1318.7
    AB-Net17.4
    Table 6. Analysis of generalization ability data on Middlebury 2014 dataset
    Jingjing ZHANG, Xingzhuo DU, Shuai ZHI, Guopeng DING. Atrous convolution and Bilateral grid network[J]. Optics and Precision Engineering, 2024, 32(3): 445
    Download Citation