Road Network Generation with City Block Attributes Using Link Attribute Aggregation

We propose a method of generating road networks with city block attributes using an encoder-decoder network and link attribute aggregation. The encoder encodes the structure of surrounding local road patterns, which enables incremental road network generation conditioned on existing road network patterns. Our contributions are threefold. First, we extend the encoder-decoder network to enable it to generate both road network graphs and the attribute of city blocks enclosed by roads. The generated links hold the attribute of adjacent city blocks as link attributes, and the aggregation of the surrounding link attributes determines the attribute of the city block surrounded by the links. Second, we propose new similarity measures and extensively evaluate the range of the similarity between actual and generated road networks. Third, we evaluate the effect of the method to integrate features of multiple paths in road networks on the similarity. We demonstrate that the proposed method can generate diverse road networks including the reproduction of actual road networks with high precision.