Academic Project: Urban Revitalization

Duration: July 2023
Location: Historical District, Guangzhou, China
Instructors: Qiaoming Deng, Kai Hu
Type: Teamwork (Main responsibilities for the following Part 2, 3, 4; Participation in Part 5, 6)
Skills: K-means Clustering, Generative Adversarial Network (GAN), Stable Diffusion & Midjourney

1. Characteristics of Guangzhou’s West Gate in the 1980s:
   • Transition to Economic Focus: In the 1980s, Guangzhou shifted its focus to economic development, witnessing increased trade and complex urban functions.
   • Architectural Diversity: Notable landmarks like White Swan Hotel and Guangzhou Garden Hotel showcased varied architectural styles, influenced by Western modernism and showcasing distinctive regional features.
   • Street Culture: Streets adorned with iconic arcade structures and bustling activities reflected the spirit and vitality of Guangzhou’s street culture.
2. Urban Typology Analysis Based on Images:
   • Utilizing K-Means Clustering: Employing elbow and silhouette methods, combined with satellite imagery, identified four urban types - traditional residential-commercial, traditional public, new commercial, and mixed-use areas.
3. Analysis of the Current State of Redevelopment Site:
   • Selected Site: The redevelopment focus is on the southern side of People’s South Historical and Cultural Street, along Jiangxi Road.
   • Urban Fabric: Recognizing the unique urban fabric of the area with open spaces, traditional arcade structures, and a mix of historic and modern buildings.
   • Design Goals: Addressing large-scale commercial architecture concerns, enhancing historical-cultural imagery, and creating a distinctive skyline.
4. Urban Morphology Generation Using GAN:
   • Preservation Strategy: Retaining existing heritage structures per conservation plans.
   • Model Selection: Employing GAN to generate a traditional public street district, with fine-tuning based on generated results and current skyline.
5. Street Design Using Stable Diffusion & Midjourney:
   • Integrating Styles: Blending neoclassical, arcade features, and modern elements for a harmonious riverside aesthetic.
   • Keyword Integration: Incorporating keywords such as 1980s, neoclassical, arcade, and modern architecture into the GPT model for street design.
6. Conclusion and Future Prospects:
   • Quantifying Urban Form: Highlighting challenges in data availability for quantifying urban form and expressing the need for improved data sources for better clustering results.
   • AI-Assisted Design Reflections: Acknowledging the efficiency and innovation benefits of AI-assisted design while recognizing potential challenges like dependency and data privacy.
   • Future Aspirations: Hoping for real-time environmental perception and optimization post AI-generated design, enhancing augmented reality applications in architecture and urban design, and improving collaboration within design teams.