Predicting Non-functional Nodes of Floorplan via Graph Neural Network (GNN) - Restructure Graph from Bubble Diagram to Floorplan Graph

Oct 26, 2026

This paper proposes an automated floorplan generation framework using Graph Neural Networks (GNN). It addresses a common limitation where prior work conflates bubble diagrams and floorplan graphs despite their different roles in spatial versus functional representation. The method builds CubiBubble5k from CubiCasa5k and CubiGraph5k, trains one GNN to convert functional lists into bubble diagrams, and a second GNN to insert non-functional spaces into functional bubbles to obtain floorplan graphs. It evaluates both models, integrates with graph2plan for real-world floorplans, and validates feasibility through case studies with visualizations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Floorplan and Graph

Bubble Diagram and Floorplan Graph

Graph Neural Network

Figure 1: Interconversion of Four Data Types

Figure 2: Research Framework

Predicting Non-functional Nodes of Floorplan via Graph Neural Network (GNN) Restructure Graph from Bubble Diagram to Floorplan Graph Zidong Liu1, Han Li2, Daniel Koehler3, Yan Li4 1,3The University of Texas at Austin 2 University of Pennsylvania 4 University of Sydney, [1](1 zidong.liu.22@utexas.edu2lihan1@upenn.edu3daniel.koehler@utexas.edu)[ zidong.liu.22@utexas.edu](1 zidong.liu.22@utexas.edu2lihan1@upenn.edu3daniel.koehler@utexas.edu)[2](1 zidong.liu.22@utexas.edu2lihan1@upenn.edu3daniel.koehler@utexas.edu)[lihan1@upenn.edu](1 zidong.liu.22@utexas.edu2lihan1@upenn.edu3daniel.koehler@utexas.edu)[3](1 zidong.liu.22@utexas.edu2lihan1@upenn.edu3daniel.koehler@utexas.edu)[daniel.koehler@utexas.edu](1 zidong.liu.22@utexas.edu2lihan1@upenn.edu3daniel.koehler@utexas.edu) [4](4yali3816@uni.sydney.edu.au)[yali3816@uni.sydney.edu.au](4yali3816@uni.sydney.edu.au) This paper presents an innovative approach to automating the floorplan generation process by employing Graph Neural Networks (GNN) to facilitate the transition from functional assignment lists to bubble diagrams and eventually to floorplan graphs. In recent years, there have been many studies on the interconversion of floorplan graph and layout design. However, these studies usually mix up floorplan graph and bubble diagram, despite their distinct roles in representing spatial and functional relationships, respectively. To address this disparity, we introduce a research framework comprising three main steps. First, we generate the CubiBubble5k dataset, which encompasses bubble diagrams and functional lists, drawing on the existing CubiCasa5k and CubiGraph5k datasets. Next, we train a GNN to transform design assignments into structured graph data, utilizing functional lists as input and bubble diagrams as output. Subsequently, we train another GNN that predictsand inserts non-functional spaces, such as corridors and anterooms, into purely functional bubble diagrams, using bubble graphs as input and floorplan graphs as output. We assess the performance of both GNNs and, by integrating our framework with the established graph2plan study, successfully demonstrate the generation of real-world floorplans from project task books. Lastly, we conduct case studies to validate the feasibility of our proposed framework. We use the existing graph2plan platform to visualize the impact of our algorithm on the final layout. Keywords: Floorplan Automation, Bubble Diagram, Graph Restructure, Graph Neural Network. INTRODUCTION Graph, or spatial networks, have been used for quantitative analysis and generative design of built environments for decades. In the 1840s in London, the government took a series of measures to establish a modern postal service, which is regarded as “one of the first forms of computable spatial networks”(Bottazzi, 2018) . Similarly, architects have been using bubble diagrams to represent the spatial organization of architectural layout designs since the 1930s (Emmons, 2006) . Floorplan and Graph In recent years, the use of graphs, networks, and related concepts has become increasingly popular in the architectural and urban studies fields. By searching for keywords such as network, graph, or pattern in architectural publications from 1990 to Volume 2 – Digital Design Reconsidered – eCAADe 41 | 529 2021, it is evident that the number of publications in the past five years has increased by over 50%(Li, Gao and Lin, 2022) . For a long time, the research of floorplan automation algorithms has been dominated by rulebased parametric ideas. Magnetizing Floorplan Generator (MagnetizingFPG ) is one of the representative studies (Egor et al., 2020) . This platform allows users to obtain an automated floor plan by entering basic parameters such as room functions, areas, and connections. Some researchers regard a floorplan as a polygon and use graph theory for floorplan automation (Shekhawat et al., 2021) . Some researchers focus on the hierarchies of gra

FAQs

What gap do the paper’s authors address in existing floorplan graph research?

They argue that prior studies often treat floorplan graphs and bubble diagrams as the same object, even though they represent different relationships and serve different purposes.

What are the two GNN components in the proposed framework?

The first GNN transforms functional assignment lists into structured graph data using bubble diagrams as output. The second GNN predicts and inserts non-functional spaces (e.g., corridors and anterooms) into functional bubble diagrams to produce floorplan graphs.

How is the CubiBubble5k dataset constructed and used?

CubiBubble5k is generated to include bubble diagrams and functional lists by drawing on CubiCasa5k and CubiGraph5k, then used to train the GNN that performs the list-to-bubble conversion.

Oct 26, 2026

Upload by FajarTechTalk