Fire Image Generation Based on ACGAN

In order to solve the problem that it is difficult to obtain fire image data in CNN training, this paper discusses the method of generating fire image by means of generative adversarial networks. How to generate the desired fire image according to the known observation variables is discussed. According to the structure of InfoGAN and ACGAN, a GAN structure for generating fire image is proposed. Fire area is selected as a known observation variable to generate the corresponding fire image. Experiments show that the network structure can generate the required images according to the values of a observed variables. And the quality of the generated image is related to the distribution of observed variables in the data set.