Article: Generative adversarial network and parametric design algorithm for public art form generation Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.68 pp.46 - 69 Abstract: In response to the challenges faced by public art design in the urbanisation process, such as low efficiency and insufficient diversity, as well as the issues of black-boxing and poor controllability inherent in traditional generative adversarial networks (GANs), this paper proposes an innovative algorithm that integrates GANs with parametric design. The algorithm aims to decouple structure and style through a dual-branch generator, achieve dynamic regulation through a parametric attention fusion module, and enhance authenticity with a multi-scale discriminator. The dual-branch generator architecture is employed to separately process structural primitives and stylistic details of the form, while introducing a parametric attention fusion module to dynamically modulate the feature fusion process. Additionally, a multi-scale discriminator and decoupling loss function are incorporated to improve generation quality and stability. Experimental results show a FID of 15.2, an inception score (IS) of 8.9, a peak signal-to-noise ratio (PSNR) of 28.5 dB, a robustness ΔFID of only 12.5%, and a user rating of 4.3 ± 0.4 for artistic value, confirming the algorithm's advantages in generation quality and practicality. This study provides an end-to-end framework, laying a technical foundation for the development of intelligent design tools. Inderscience Publishers - linking academia, business and industry through research

Title: Generative adversarial network and parametric design algorithm for public art form generation

Authors: Qian Ye

Addresses: School of Drawing, Hubei Institute of Fine Arts, Wuhan, Hubei 430205, China

Abstract: In response to the challenges faced by public art design in the urbanisation process, such as low efficiency and insufficient diversity, as well as the issues of black-boxing and poor controllability inherent in traditional generative adversarial networks (GANs), this paper proposes an innovative algorithm that integrates GANs with parametric design. The algorithm aims to decouple structure and style through a dual-branch generator, achieve dynamic regulation through a parametric attention fusion module, and enhance authenticity with a multi-scale discriminator. The dual-branch generator architecture is employed to separately process structural primitives and stylistic details of the form, while introducing a parametric attention fusion module to dynamically modulate the feature fusion process. Additionally, a multi-scale discriminator and decoupling loss function are incorporated to improve generation quality and stability. Experimental results show a FID of 15.2, an inception score (IS) of 8.9, a peak signal-to-noise ratio (PSNR) of 28.5 dB, a robustness ΔFID of only 12.5%, and a user rating of 4.3 ± 0.4 for artistic value, confirming the algorithm's advantages in generation quality and practicality. This study provides an end-to-end framework, laying a technical foundation for the development of intelligent design tools.

Keywords: generative adversarial network; GAN; public art; generation algorithm; parametric design; double branch generator.

DOI: 10.1504/IJICT.2026.154350

International Journal of Information and Communication Technology, 2026 Vol.27 No.68, pp.46 - 69

Received: 09 Dec 2025
Accepted: 28 Feb 2026
Published online: 23 Jun 2026 *

Title: Generative adversarial network and parametric design algorithm for public art form generation

Keep up-to-date