Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Boundary-Constrained Diffusion Models for Floorplan Generation: Balancing Realism and Diversity

Leonardo Stoppani, Davide Bacciu, Shahab Mokarizadeh

TL;DR

This work tackles the tension between realism and diversity in floorplan generation by introducing Boundary Cross-Attention (BCA) to enforce boundary constraints, and a Diversity Score (DS) to quantify layout variation under fixed conditioning. Built as an extension of HouseDiffusion, the approach enables explicit boundary conditioning via $B$ and a controllable guidance scale $\lambda$ to trade adherence for creativity, while DS provides a direct measure of diversity not captured by $FID$. An Out-of-Distribution (OOD) evaluation reveals reliance on dataset priors and limited generalization to novel geometries, underscoring the need for models that balance fidelity, diversity, and generalization. Overall, the proposed methods advance controllable, boundary-aware, and more robust floorplan generation, with practical implications for architectural design and planning.

Abstract

Diffusion models have become widely popular for automated floorplan generation, producing highly realistic layouts conditioned on user-defined constraints. However, optimizing for perceptual metrics such as the Fréchet Inception Distance (FID) causes limited design diversity. To address this, we propose the Diversity Score (DS), a metric that quantifies layout diversity under fixed constraints. Moreover, to improve geometric consistency, we introduce a Boundary Cross-Attention (BCA) module that enables conditioning on building boundaries. Our experiments show that BCA significantly improves boundary adherence, while prolonged training drives diversity collapse undiagnosed by FID, revealing a critical trade-off between realism and diversity. Out-Of-Distribution evaluations further demonstrate the models' reliance on dataset priors, emphasizing the need for generative systems that explicitly balance fidelity, diversity, and generalization in architectural design tasks.

Boundary-Constrained Diffusion Models for Floorplan Generation: Balancing Realism and Diversity

TL;DR

This work tackles the tension between realism and diversity in floorplan generation by introducing Boundary Cross-Attention (BCA) to enforce boundary constraints, and a Diversity Score (DS) to quantify layout variation under fixed conditioning. Built as an extension of HouseDiffusion, the approach enables explicit boundary conditioning via and a controllable guidance scale to trade adherence for creativity, while DS provides a direct measure of diversity not captured by . An Out-of-Distribution (OOD) evaluation reveals reliance on dataset priors and limited generalization to novel geometries, underscoring the need for models that balance fidelity, diversity, and generalization. Overall, the proposed methods advance controllable, boundary-aware, and more robust floorplan generation, with practical implications for architectural design and planning.

Abstract

Diffusion models have become widely popular for automated floorplan generation, producing highly realistic layouts conditioned on user-defined constraints. However, optimizing for perceptual metrics such as the Fréchet Inception Distance (FID) causes limited design diversity. To address this, we propose the Diversity Score (DS), a metric that quantifies layout diversity under fixed constraints. Moreover, to improve geometric consistency, we introduce a Boundary Cross-Attention (BCA) module that enables conditioning on building boundaries. Our experiments show that BCA significantly improves boundary adherence, while prolonged training drives diversity collapse undiagnosed by FID, revealing a critical trade-off between realism and diversity. Out-Of-Distribution evaluations further demonstrate the models' reliance on dataset priors, emphasizing the need for generative systems that explicitly balance fidelity, diversity, and generalization in architectural design tasks.
Paper Structure (9 sections, 2 equations, 2 figures, 3 tables)

This paper contains 9 sections, 2 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Boundary-Constrained Vector Floorplan Generation
  3. Boundary Cross-Attention Module.
  4. Diversity Score.
  5. Experiments
  6. Experimental Setup.
  7. Model Performance Comparison.
  8. OOD Evaluation.
  9. Conclusions

Figures (2)

  • Figure 1: Our HouseDiffusion (HD) framework extended to include boundary conditioning.
  • Figure 2: Example of failure due to the model's reliance on the training data distribution for the number of corners of each room.