Abstract
Floor plans depict building layouts and are often represented as graphs tocapture the underlying spatial relationships. Comparison of these graphs iscritical for applications like search, clustering, and data visualization. Themost successful methods to compare graphs \ie, graph matching networks, rely oncostly intermediate cross-graph node-level interactions, therefore being slowin inference time. We introduce \textbf{LayoutGKN}, a more efficient approachthat postpones the cross-graph node-level interactions to the end of the jointembedding architecture. We do so by using a differentiable graph kernel as adistance function on the final learned node-level embeddings. We show thatLayoutGKN computes similarity comparably or better than graph matching networkswhile significantly increasing the speed.\href{https://github.com/caspervanengelenburg/LayoutGKN}{Code and data} areopen.