A StyleGAN2-based framework for generating and evaluating urban color landscapes and street vitality

Urban color landscapes play a significant role in shaping perceptual experience and street vitality. We proposed a style-based generative adversarial network 2 (StyleGAN2)-inspired generative framework for creating urban colorscapes and quantitatively assessing their vitality. The approach integrated advanced data preprocessing, generator–discriminator architecture, and hyperparameter optimization using a non-saturating logistic loss function. Vitality was evaluated through three chromatic indicators—saturation, contrast, and diversity—and validated against behavioral (pedestrian volume) and socioeconomic (point-of-interest density) data via correlation analysis (r = 0.47–0.68). The model achieved theoretical convergence (Fréchet Inception Distance < 15) and optimality, while ablation experiments with a deep convolutional GAN, Wasserstein GAN with gradient penalty, and StyleGAN3 confirmed its superior generative performance. The synthesized images exhibited an 18.2% increase in saturation and a 10.5% increase in diversity relative to real-world scenes, suggesting a strong positive association with urban vitality, as established in our correlation analysis. Computational efficiency was enhanced through mixed-precision training, reducing total processing time. Empirical and perceptual validations confirmed the framework’s robustness, offering a reproducible pathway for artificial intelligence-driven urban color planning.

Text 126478.pdf - Published Version Available under License Creative Commons Attribution. Download (4MB) Official URL or Download Paper: https://accscience.com/journal/IJOCTA/16/1/10.3692...

Actions (login required)

View Item