Style Transfer of Urban Road Images Using Generative Adversarial Networks With Structural Details

To evaluate the driving behavior of unmanned vehicles, the testing of driving algorithms using urban road images is necessary. In this article, we propose a framework using generative adversarial networks (GANs) with structural information for image style transfer: StructureGAN and GradientGAN. Different types of urban image transfers are generated using the proposed framework, such as day to night, sunny to foggy, and summer to winter transfers. The proposed method can well maintain the integrity of foreground objects and the image structural information. Artifacts such as image distortion and foreground disappearance are eliminated. The experiments with the baseline methods indicate the effectiveness of the proposed framework, which can produce transferred images with high quality.