• Acta Optica Sinica
  • Vol. 40, Issue 18, 1810001 (2020)
Ruiming Jia1、*, Tong Li1, Shengjie Liu1, Jiali Cui1, and Fei Yuan2
Author Affiliations
  • 1School of Information Science and Technology, North China University of Technology, Beijing 100144, China
  • 2Digital Content Technology and Media Service Research Center, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China
  • show less
    DOI: 10.3788/AOS202040.1810001 Cite this Article Set citation alerts
    Ruiming Jia, Tong Li, Shengjie Liu, Jiali Cui, Fei Yuan. Infrared Simulation Based on Cascade Multi-Scale Information Fusion Adversarial Network[J]. Acta Optica Sinica, 2020, 40(18): 1810001 Copy Citation Text show less
    Proposed network of CMIF-GAN
    Fig. 1. Proposed network of CMIF-GAN
    Proposed network of G2
    Fig. 2. Proposed network of G2
    Multi-scale fusion module
    Fig. 3. Multi-scale fusion module
    Proposed network of Gg
    Fig. 4. Proposed network of Gg
    Infrared simulation images generated by different algorithms. (a) Car; (b) bicycle rider; (c) building; (d) long-distance vehicle
    Fig. 5. Infrared simulation images generated by different algorithms. (a) Car; (b) bicycle rider; (c) building; (d) long-distance vehicle
    Results comparison of first level network and CMIF-GAN
    Fig. 6. Results comparison of first level network and CMIF-GAN
    Encoder (G1/Gs filters)Number of channels G1/GsDecoder (G1/Gsfilters)Number of channels G1/Gs
    dconv16 (3/3)3/3
    conv1 (64/4)64/4dconv15 (64/4)128/8
    conv2 (128/8)128/8dconv14 (128/8)256/16
    conv3 (256/16)256/16dconv13 (256/16)512/32
    conv4 (512/32)512/32dconv12 (512/32)1024/64
    conv5 (512/32)512/32dconv11 (512/32)1024/64
    conv6 (512/32)512/32dconv10 (512/32)1024/64
    conv7 (512/32)512/32dconv9 (512/32)1024/64
    conv8 (512/32)512/32
    Table 1. Detailed configuration about G1 and Gs
    NetworkInput/outputchannelsStrideBNActivationfunction
    conv16/642NL
    conv2641282YL
    conv3128/2562YL
    conv4256/5122YL
    conv5512/5121YL
    conv6512/11NS
    Table 2. Detailed configuration about discriminator
    MethodThe lower, the betterThe higher, the better
    RelAvg log10RMSδ<1.25PSNRSSIM
    FCRN[11]0.2860.1441.0600.40921.2040.962
    FLED-Net[12]0.2380.1000.8530.60222.9210.987
    Pix2pix[20]0.2480.1070.9060.57122.4310.985
    Selection-GAN[23]0.2840.1120.9580.55421.9760.982
    Proposed0.2570.1020.8760.61222.6570.989
    Table 3. Comparison of objective indicators of algorithms
    MethodThe lower, the betterThe higher, the better
    RelAvg log10RMSδ<1.25PSNRSSIM
    First level0.2650.1070.9180.58922.3100.987
    Proposed0.2570.1020.8760.61222.6570.989
    Table 4. Comparison of first level network and CMIF-GAN
    SetupThe lower, the betterThe higher, the better
    RelAvg log10RMSδ<1.25PSNRSSIM
    -Gs,Gr0.2680.1070.9120.58622.3210.988
    -Gg0.2760.1060.9250.59422.2800.986
    -Gs0.2640.1050.9010.59822.5100.987
    Proposed0.2570.1020.8760.61222.6570.989
    Table 5. Comparison of auxiliary tasks experiments
    SetupThe lower, the betterThe higher, the better
    RelAvg log10RMSδ<1.25PSNRSSIM
    -MFM0.2680.1050.9050.60022.4650.987
    Proposed0.2570.1020.8760.61222.6570.989
    Table 6. Comparison of MFM module experiments
    Ruiming Jia, Tong Li, Shengjie Liu, Jiali Cui, Fei Yuan. Infrared Simulation Based on Cascade Multi-Scale Information Fusion Adversarial Network[J]. Acta Optica Sinica, 2020, 40(18): 1810001
    Download Citation