Traffic thermal infrared texture generation based on siamese semantic CycleGAN

•A novel thermal infrared texture generation algorithm is proposed to transfer visible images into thermal infrared images, by generative adversarial networks (GAN). Different from traditional thermal simulation procedure, the proposed SS-CycleGAN could generate infrared images with both thermal textures and clear visible edge details, with no extra environment information.•Dual discriminator is introduced to balance the frequency performance of generated thermal infrared images.•A novel siamese semantic loss is designed to enhance the inner connection between visible and thermal infrared images, by associating the two different procedure of CycleGAN.