Graph Neural Network-Based Clustering of Floor Plans via Visual Segments

Mid-level segments of spatial articulation exist between individual points and the spatial whole. Based on visibility graphs, these can be defined as segments analogous to graph-theoretic communities. This paper proposes a method for clustering such segments using Graph Neural Network-based Invariant Information Clustering. Clustering is performed on both individual segments and the entire graph as an aggregation of these segments. Consequently, the segment-level clustering focuses on the connectivity and relative positioning between segments, while the graph-level clustering captures the specific features and distribution patterns of the segments within each graph.