Geração de Loteamentos Residenciais com Deep Learning baseada em Redes Adversárias Generativas

• Milena Lucas dos Santos Itaipu Parquetec
• Rafhael Pereira Zrenner Itaipu Parquetec
• Fabiana Frata Furlan Peres USP / UNIOESTE https://orcid.org/0000-0002-3937-3432
• Érick Oliveira Rodrigues UTFPR

DOI: https://doi.org/10.5753/latinoware.2025.16564

Resumo

Este trabalho apresenta um estudo exploratório do uso de redes generativas Pix2Pix para apoiar a equipe de Arquitetura e Engenharia do Itaipu Parquetec na geração automática de layouts preliminares de loteamentos habitacionais. A proposta busca auxiliar o planejamento urbano inicial por meio da criação de layouts a partir de parâmetros definidos, como dimensões de quadras, lotes e terrenos. Para isso, foi desenvolvido um conjunto de dados com 700 amostras, construído a partir de imagens geradas no AutoCAD em combinação com mapas reais da malha urbana de Foz do Iguaçu, Brasil. A metodologia envolveu o pré-processamento dos dados, a geração da malha viária (ruas), o treinamento do modelo de Redes Adversárias Generativas (GANs) condicional e o ajuste de hiperparâmetros. A avaliação quantitativa, realizada com o Fréchet Inception Distance (FID), indicou valores de 40,61 para cenários simples e 73,09 após a inclusão da malha viária, confirmando o aumento da complexidade da tarefa. Ainda assim, a abordagem pode ser usada como ferramenta de apoio nas etapas iniciais do planejamento urbano. Como trabalhos futuros, pretende-se ampliar a base de dados para diferentes tipologias urbanas, disponibilizá-la como benchmark para a área e investigar o uso de modelos de Difusão como alternativa para melhorar a fidelidade geométrica dos layouts gerados.

Palavras-chave: Redes Adversárias Generativas, Pix2Pix, Design Generativo, Layout Residencial

Referências

S. Chaillou, “Archigan: Artificial intelligence x architecture,” in Architectural Intelligence: Selected Papers from the 1st International Conference on Computational Design and Robotic Fabrication (CDRF 2019). Springer, 2020, pp. 117–127. [Online]. DOI: 10.1007/978-981-15-6568-7_8

P. Sun, F. Yan, Q. He, and H. Liu, “The development of an experimental framework to explore the generative design preference of a machine learning-assisted residential site plan layout,” Land, vol. 12, no. 9, p. 1776, 2023.

M. Rahbar, M. Mahdavinejad, M. Bemanian, A. H. D. Markazi, and L. Hovestadt, “Generating synthetic space allocation probability layouts based on trained conditional-gans,” Applied Artificial Intelligence, vol. 33, no. 8, pp. 689–705, 2019.

R. Tian, “Suggestive site planning with conditional gan and urban gis data,” in Proc. 2020 DigitalFUTURES (CDRF), P. F. Y. et al., Ed., 2021, pp. 103–113.

A. Chen, “Generation of layouts for living spaces using conditional generative adversarial networks: Designing floor plans that respect both a boundary and high-level requirements,” arXiv preprint, 2022. [Online]. Available: [link]

N. N. et al., “House-gan++: Generative adversarial layout refinement network towards intelligent computational agent for professional architects,” in Proc. IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 13 632–13 641.

——, “House-gan: Relational generative adversarial networks for graphconstrained house layout generation,” in Proc. Eur. Conf. on Computer Vision (ECCV), vol. 12346. Springer, 2020, pp. 162–177.

R. C. et al., “Generative design of outdoor green spaces based on generative adversarial networks,” Buildings, vol. 13, no. 4, p. 1083, 2023.

J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems, vol. 33, pp. 6840–6851, 2020.

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems, vol. 30, 2017.

K. Xu, Y. Li, and H. Wang, “Urbangenai: Reconstructing urban landscapes using panoptic segmentation and diffusion models,” arXiv preprint, 2024. [Online]. Available: [link]

J. Smith, L. Chen, and M. Zhao, “Generating urban road networks with conditional diffusion models,” ISPRS International Journal of Geo-Information, vol. 13, no. 6, p. 203, 2024. [Online]. Available: [link]

A. Li, P. Zhang, and R. Wang, “Conditional diffusion model for urban morphology prediction,” Remote Sensing, vol. 16, no. 10, p. 1799, 2024. [Online]. Available: [link]

Z. Qian, M. Li, and R. Zhang, “Dalle-urban: Capturing the urban design expertise of large text-to-image transformers,” arXiv preprint, 2022. [Online]. Available: [link]

J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” arXiv preprint, 2020. [Online]. Available: [link]

P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, “Image-to-image translation with conditional adversarial networks,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2017, pp. 5967–5976.

D. Hölscher, C. Reich, F. Gut, M. Knahl, and N. L. Clarke, “Pix2pix hyperparameter optimisation prediction,” Procedia Computer Science, vol. 225, pp. 1009–1018, 2023.

J.-Y. Zhu, “Can you please describe about the gan mode in details?” GitHub issue #1011, pytorch-CycleGAN-and-pix2pix, 2020, accessed: 2025-05-20. [Online]. Available: [link]

M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,” in Advances in Neural Information Processing Systems, vol. 30, 2017, pp. 6626–6637.

C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the inception architecture for computer vision,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 2818–2826.

T. Oanda, “A review of the image quality metrics used in image generative models,” Paperspace Blog, 2023, accessed: 2025-05-07. [Online]. Available: [link]