Fig. 5

Street network images (top row) with most similar street networks (rows below) based on the Euclidean distance between their urban network vectors. The latent representations, obtained using the trained encoder, seem to capture well network properties such as density, orientation or road shape