Urban layouts are incredibly diverse and complex. Simple metrics like building footprint area are incomplete in describing the higher-dimensional complexity and spatial hierarchy of these layouts. Other design language, like 'urban grain' or 'morphology', are qualitative. Consensus on describing urban layouts or classifying them into meaningful urban typologies is hard to come by.
This project is an attempt to describe urban layouts in a standardized, objective way by encoding their spatial structure using deep neural networks. We show how their encoded latent features can be used in downstream urban planning applications, including neigborhood analysis and generating composite urban layouts.
Check out the webapp here: https://share.streamlit.io/lazfishing/manhattan-city-layouts