Laser & Optoelectronics Progress, Volume. 60, Issue 12, 1210008(2023)

fire-GAN: Flame Image Generation Algorithm Based on Generative Adversarial Network

Kui Qin, Xinguo Hou*, Feng Zhou, Zhengjun Yan, and Leping Bu
Author Affiliations
  • School of Electrical Engineering, Naval University of Engineering, Wuhan 430033, Hubei, China
  • show less
    Figures & Tables(15)
    Example of RGB-uv histogram
    Difference between HistoGAN and styleGAN2
    Example of flame image segmentation using Equ. (5)
    Example of generating flame image using HistoGAN
    Comparison of flame effect generated by two parts of datasets
    Comparison of flame generation effects under different conditions of roundness loss function
    Comparison of flame image generated by fire-GAN and MixNMatch
    Comparison of image effects generated by GN, APA, and fire-GAN
    Relationship between FID of image generated by GN, APA, and fire-GAN and training times
    Comparison of flame images generated by different networks
    • Table 1. Roundness of flames and disturbances

      View table

      Table 1. Roundness of flames and disturbances

      ParameterFlameFlashlightCar lightsSunlight through windows
      C0.2790.7140.6660.585
    • Table 2. Quantitative evaluation of flame image generated by two parts of datasets

      View table

      Table 2. Quantitative evaluation of flame image generated by two parts of datasets

      DatasetFIDIS
      Without flame segmentation80.092.39
      With flame segmentation59.232.81
    • Table 3. Comparison of average values of flame roundness in images generated under different conditions of roundness loss function

      View table

      Table 3. Comparison of average values of flame roundness in images generated under different conditions of roundness loss function

      ParameterWithout lossRd_lossRg_lossRd_loss+Rg_lossReal image
      C0.4340.3790.3300.3190.279
    • Table 4. Comparison of R, G, and B mean values of images generated by fire-GAN and MixNMatch

      View table

      Table 4. Comparison of R, G, and B mean values of images generated by fire-GAN and MixNMatch

      NetworkRGB
      Target image233.6397217.8905201.2066
      fire-GAN228.3991213.4572194.0862
      MixNMatch204.1542203.5226171.0034
    • Table 5. Quantitative evaluation of different networks

      View table

      Table 5. Quantitative evaluation of different networks

      ParameterGANSAGANMixNMatchcontent-aware-GANstyleGAN2fire-GAN
      FID129.02125.35140.3468.8560.5259.23
      IS2.062.001.912.762.792.81
    Tools

    Get Citation

    Copy Citation Text

    Kui Qin, Xinguo Hou, Feng Zhou, Zhengjun Yan, Leping Bu. fire-GAN: Flame Image Generation Algorithm Based on Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2023, 60(12): 1210008

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Mar. 14, 2022

    Accepted: Jun. 13, 2022

    Published Online: May. 23, 2023

    The Author Email: Hou Xinguo (hxinguo2008@126.com)

    DOI:10.3788/LOP220989

    Topics