In the field of architecture and urban planning, the isovist theory is used for evaluating spaces. In this theory, metrics, such as area or edge length, are employed to simplify higher-dimension isovist volumes. In this study, we propose a visibility evaluation method using a deep neural network as a feature extractor that extracts features from isovist point clouds. A classification and clustering network were tested by evaluating five architectures. The results show that the networks can extract valuable features and analyze the visibility using architectural characters, spatial spread, their direction, etc.
