Floor Plan Reconstruction from Sparse Views: Combining Graph Neural Network with Constrained Diffusion

We address the challenging problem of floor plan reconstruction from sparse views and a room-connectivity graph. As a first stage, we construct a flexible graph-structure unifying the connectivity graph and the sparse observed data. Using our Graph Neural Network architecture, we can then refine the available information and predict unobserved room properties. In a second step, we introduce a Constrained Diffusion Model to reconstruct consistent floor plan matching the available information, despite of its sparsity. More precisely, we use a Cross-Attention mechanism armed with shape descriptors to guarantee that the generated floor plan reflects both the input room connectivity and the geometry observed in the sparse views.

1. Introduction

Floor plans serve as a common representation to convey accurate information about the layout of houses and apartments, facilitating effective communication between architects, designers, or parties to a real estate transaction. When considering real-case scenarios encountered in real estate applications, accurate floor plans are rarely available. Instead, apartment descriptions are often provided in the form of a small number of photos taken during visits, providing only partial information, and supplemented by textual descriptions of room types. The ability to reconstruct a plausible floor plan from such incomplete and sparse data would be of great benefit in real estate in the objective of converging toward the real plan when combined with additional data.