• Optics and Precision Engineering
  • Vol. 31, Issue 24, 3651 (2023)
Wei CAI, Bo JIANG*, Xinhao JIANG, and Zhiyong YANG
Author Affiliations
  • Armament Launch Theory and Technology Key Discipline Laboratory of PRC, Rocket Force University of Engineering, Xi′an710025, China
  • show less
    DOI: 10.37188/OPE.20233124.3651 Cite this Article
    Wei CAI, Bo JIANG, Xinhao JIANG, Zhiyong YANG. Infrared image generation with unpaired training samples[J]. Optics and Precision Engineering, 2023, 31(24): 3651 Copy Citation Text show less
    Comparison of image effects generated by two infrared simulation algorithms
    Fig. 1. Comparison of image effects generated by two infrared simulation algorithms
    Basic framework of VTIGAN
    Fig. 2. Basic framework of VTIGAN
    Network structure of generator
    Fig. 3. Network structure of generator
    Network structure of discriminator
    Fig. 4. Network structure of discriminator
    Example of experimental results
    Fig. 5. Example of experimental results
    ModuleEncoderConvertDecoder
    Convolution kernelNetwork ConnectionNumber of convolution kernelsModuleNumberConvolution kernelNetwork ConnectionNumber of convolution kernels
    Instruction7×7Conv ,BN ,ReLU64Transformer13×3ConvTranspose ,BN ,ReLU128
    3×3Conv ,BN ,ReLU1283×3ConvTranspose ,BN ,ReLU64
    3×3Conv ,BN ,ReLU2567×7Conv ,BN ,ReLU3
    Table 1. Generators internal parameters
    NamesRelated configurations
    GPUNVIDIA Quadro GV100
    CPUInter Xeon Silver 4210/128 G
    GPU Memory size32 G
    Operating systemsWin10
    Computing platformCUDA11.0
    Table 2. Experimental platform configuration
    ParameterEpochBatch sizeLearning rateλGANλMCλSSλIDT
    Value30010.000 212101
    Table 3. Setting of experimental parameters
    MethodPub’YearArchitectural SceneTraffic Scene
    PSNR/dBSSIMFIDPSNR/dBSSIMFID
    GLANet202112.7200.63175.3913.8230.68068.27
    CUTECCV’202012.8950.63471.2513.9950.67562.21
    CycleGANICCV’201713.1380.64666.5214.7790.76650.11
    DSMAPECCV’202014.3510.73850.3815.3640.78042.88
    UGATITICLR’201915.2100.76546.9216.6410.80538.57
    VTIGANOur15.8500.79639.3616.7500.81836.48
    Table 4. Comparison of image evaluation indexes
    Loss functionPSNR/dBSSIMFID
    Adversarial lossMulti-layer contrast lossStyle similarity lossIdentity loss
    15.1820.72145.84
    10.1460.51396.47
    9.8510.49498.18
    15.8500.79639.36
    Table 5. Results of ablation experiments
    Wei CAI, Bo JIANG, Xinhao JIANG, Zhiyong YANG. Infrared image generation with unpaired training samples[J]. Optics and Precision Engineering, 2023, 31(24): 3651
    Download Citation