Deterioration of pre-war and rehabilitation of post-war urbanscapes using generative adversarial networks

Çiçek S., Turhan G. D., Taşer A.

International Journal of Architectural Computing, 2023 (ESCI) identifier

  • Publication Type: Article / Article
  • Publication Date: 2023
  • Doi Number: 10.1177/14780771231181237
  • Journal Name: International Journal of Architectural Computing
  • Journal Indexes: Emerging Sources Citation Index (ESCI), Scopus, Aerospace Database, Applied Science & Technology Source, Art Source, Communication Abstracts, Compendex, Computer & Applied Sciences, Index Islamicus, INSPEC, Metadex, Civil Engineering Abstracts
  • Keywords: artificial intelligence, CycleGAN, generative adversarial network, machine learning, pix2pix GAN, Post-war, urban rehabilitation
  • Istanbul Technical University Affiliated: No


The urban built environment of contemporary cities confronts a constant risk of deterioration due to natural or artificial reasons. Especially political aggression and war conflicts have significant destructive effects on architectural and cultural heritage buildings. The post-war urbanscapes demonstrate the striking effects of the armed conflicts during the hot war encounters. However, the residues of the urbanscapes become the actual indicators of damage and loss. Since today we can make future predictions using a variety of machine learning algorithms, it is possible to represent hybrid projections of urban heterotopias. In this context, this research proposes to explore dystopian post-war projections for modern cities based on their architectural styles and demonstrate the utopian scenarios of rehabilitation possibilities for the damaged urban built environment of post-war cities by using generative adversarial network (GAN) algorithms. Two primary datasets containing the post-war and pre-war building facades have been given as the input data for the CycleGAN and pix2pix GAN models. Thus, two different image-to-image GAN models have been compared regarding their ability to produce legible building facade projections in architectural features. Besides, the machine learning process results have been discussed in terms of cities’ utopian and dystopian future predictions, demonstrating the war conflicts’ immense effects on the built environment. Moreover, the immediate consequence of the destructive aggression on tangible and intangible architectural heritage would become visible to inhabitants and policymakers when the AI-generated rehabilitation potentials have been exposed.