Generative Adversarial Network Style Transfer for Sustainable Urban Restoration Planning in Ukraine

Due to the ongoing Russian invasion in Ukraine, many Ukrainian cities received immense damage or were destroyed. Restoring these cities is set to become one of the main priorities of the Ukrainian government in the near future. Because of its importance and sensitivity, cutting-edge technology should be used at every step of the task. The restoration process should use best European experience and practices. Generative Adversarial Networks (GAN) style transfer can help with modelling how the city would look if restored using techniques from selected European cities. Moreover, it can model the properties of smart planning of city, for instance a green zones optimization. This paper presents a novel approach to urban restoration planning using GANs to capture and translate architectural and urban planning styles from European cities into the reconstruction of a Ukrainian city. The study focuses on the destroyed city of Mariupol and evaluates the impact of these stylistic influences on greenspace distribution in the restored urban environment. In our case study, we transferred architectural features of Polish and Italian urban project and determined that for this specific city, the Polish urban planning model would be more beneficiary. The total proportion of greenspaces in the area was shown to be 7.18% with Polish project, compared to 5.32% of the Italian model.