Graph2Plan: Learning Floorplan Generation from Layout Graphs

ACM Transactions on Graphics (Proceedings of SIGGRAPH 2020)


Ruizhen Hu1    Zeyu Huang1    Yuhan Tang1    Oliver Van Kaick2    Hao Zhang3    Hui Huang1*

1Shenzhen University    2Carleton University    3Simon Fraser University




Fig. 1. Our deep neural network Graph2Plan is a learning framework for automated floorplan generation from layout graphs. The trained network can generate floorplans based on an input building boundary only (a-b), like in previous works. In addition, we allow users to add a variety of constraints such as room counts (c), room connectivity (d), and other layout graph edits. Multiple generated floorplans which fulfill the input constraints are shown.



Abstract

We introduce a learning framework for automated floorplan generation which 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 component of our learning framework is a deep neural network, Graph2Plan, which converts a layout graph, along with a building boundary, into a floorplan that fulfills both the layout and boundary constraints. Given an input building boundary, we allow a user to specify room counts and other layout constraints, which are used to retrieve a set of floorplans, with their associated layout graphs, from a database. For each retrieved layout graph, along with the input boundary, Graph2Plan first generates a corresponding raster floorplan image, and then a refined set of boxes representing the rooms. Graph2Plan is trained on RPLAN, a large-scale dataset consisting of 80K annotated floorplans. The network is mainly based on convolutional processing over both the layout graph, via a graph neural network (GNN), and the input building boundary, as well as the raster floorplan images, via conventional image convolution. We demonstrate the quality and versatility of our floorplan generation framework in terms of its ability to cater to different user inputs.We conduct both qualitative and quantitative evaluations, ablation studies, and comparisons with state-of-the-art approaches.



Fig. 2. Overview of our framework for automated floorplan generation, which combines generative modeling using deep neural networks and user-in-the-loop design. (a) The user inputs a building boundary and can also specify constraints on room numbers, locations, and adjacencies. (b) We retrieve possible graph layouts from a dataset based on the user input. (c) Once graphs have been retrieved, we automatically transfer and adjust the graphs to the input boundary. The user can then interactively edit the room locations and adjacencies on the adjusted graphs. (d) The graphs guide our network in generating corresponding floorplans, encoded as a set of room bounding boxes and a floorplan raster image. (e) We post-process this output to obtain the final, vectorized floorplan.




Fig. 6. Architecture of our Graph2Plan network. The network takes as input a layout graph and building boundary, and outputs initial room bounding boxes {B0 i }, refined room boxes {B1 i }, and a raster image I of the floorplan. The processing is carried out with a graph neural network (GNN), convolutional neural network (CNN), fully-connected layers (Box), and a BoxRefineNet (detailed in Figure 7).




Fig. 11. Our interface for user-in-the-loop design of floorplans. Left: input boundary. Middle: retrieved floorplans. Right: floorplan and layout graph of a retrieved result. Please refer to the text for more details.




Fig. 12. Gallery of floorplans generated with our method. The rows show results generated for different input boundaries, while the columns show results generated for different constraints. The constraints are the desired number of three room types: bedroom (in yellow), bathroom (blue), and balcony (green). The constraints are shown on the bottom of each column.



Data & Code

Note that the DATA and CODE are free for Research and Education Use ONLY. 

Please cite our paper (add the bibtex below) if you use any part of our ALGORITHM, CODE, DATA or RESULTS in any publication.

Link: https://github.com/HanHan55/Graph2plan 


Acknowledgement

We thank the reviewers for their valuable comments and Freepik, Shutterstock for high-quality infographic designs. This work is supported in parts by NSFC (61761146002, 61602310, 61802265), Guangdong Provincial Natural Science Foundation (2018A030310426), Shenzhen Innovation Program (JCYJ20170302154106666), LHTD (20170003), the National Engineering Laboratory for Big Data System Computing Technology, and the Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ).We thank the anonymous reviewers for their valuable comments. This work was supported in part by NSFC (61872250, 61861130365, 61761146002), GD Higher Education Key Program (2018KZDXM058), GD Science and Technology Program (2015A030312015), LHTD (20170003), NSERC Canada (611370, 611649, 2015-05407), gift funds from Adobe, National Engineering Laboratory for Big Data System Computing Technology, and Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ), Shenzhen University.


Bibtex
@article{Graph2Plan20,
title = {
Graph2Plan: Learning Floorplan Generation from Layout Graphs},
author = {Ruizhen Hu and Zeyu Huang and Yuhan Tang and Oliver Van Kaick and Hao Zhang and Hui Huang},
journal = {ACM Transactions on Graphics (Proceedings of SIGGRAPH 2020)},
volume = {39},
number = {4},
pages = {118:1--118:14},  
year = {2020},



Downloads(faster for people in China)

Downloads(faster for people in other places)