End-to-End Graph-Constrained Vectorized Floorplan Generation with Panoptic Refinement

AbstractThe automatic generation of floorplans given user inputs has great potential in architectural design and recently been explored the computer vision community. However, majority existing methods synthesize format rasterized images, which are difficult to edit or customize. In this paper, we aim as sequences 1-D vectors, eases interaction customization. To generate high fidelity vectorized floorplans, propose a novel two-stage framework, including draft stage multi-round refining stage. first stage, encode room connectivity graph input by users with convolutional network (GCN), then apply an autoregressive transformer initial floorplan sequence. polish more visually appealing further panoptic refinement (PRN) composed GCN network. The PRN takes generated sequence refines while encouraging correct our proposed geometric loss. We have conducted extensive experiments on real-world dataset, results show that method achieves state-of-the-art performance under different settings evaluation metrics.