Style Transfer of Computer-Generated Orthophoto Landscape Images Into a Realistic Look

In this paper, we present a novel application of Neural Style Transfer (NST) for converting procedurally generated orthophoto landscape images into highly realistic representations. Traditionally used to apply artistic styles to images, we adapt NST to transfer the photorealistic qualities of real aerial orthophoto images to synthetic terrain images. This enables the creation of realistic visuals from generated landscapes, addressing the common issues of stylization, abstraction, and inaccuracy in synthetic imagery. Our approach involves upgrading and modifying existing NST techniques and their comparison. The evaluation demonstrates that our methods produce more convincing and realistic results than general generative models. These findings highlight the potential of NST in enhancing the realism of computer-generated landscapes, with possible applications in urban planning, environmental simulations, video games, and the film industry.