Generation and Control of Color Flame Image Based on GAN

In order to improve the shortcomings of current flame image generation methods based on coding control, this paper proposes to describe the flame shape by the flame height and width, and generate corresponding color rectangles in advance according to the flame height and width. This approach converts the encoding into a more informative graph as a conditional input to the generator. The experiment shows that the flame height and width can be improved effectively by controlling the generated flame in graphic form. The IOU of flame achieves 0.906 and the authenticity achieves 91.4%.