Graph2Plan: Learning Floorplan Generation from Layout Graphs

摘要

We introduce a learning framework for automated floorplan generationwhich combines generative modeling using deep neural networks and userin-the-loop designs to enable human users to provide sparse design constraints.Such constraints are represented by a layout graph. The core componentof our learning framework is a deep neural network, Graph2Plan,which converts a layout graph, along with a building boundary, into a floorplanthat fulfills both the layout and boundary constraints. Given an inputbuilding boundary, we allow a user to specify room counts and other layoutconstraints, which are used to retrieve a set of floorplans, with theirassociated layout graphs, from a database. For each retrieved layout graph,along with the input boundary, Graph2Plan first generates a correspondingraster floorplan image, and then a refined set of boxes representing therooms. Graph2Plan is trained on RPLAN, a large-scale dataset consisting of 80K annotated floorplans. The network is mainly based on convolutionalprocessing over both the layout graph, via a graph neural network (GNN),and the input building boundary, as well as the raster floorplan images, viaconventional image convolution. We demonstrate the quality and versatilityof our floorplan generation framework in terms of its ability to cater to differentuser inputs.We conduct both qualitative and quantitative evaluations,ablation studies, and comparisons with state-of-the-art approaches.