Shih, Chia-YingPeng, Chi-HanCabiddu, DanielaSchneider, TeseoAllegra, DarioCatalano, Chiara EvaCherchi, GianmarcoScateni, Riccardo2022-11-082022-11-082022978-3-03868-191-52617-4855https://doi.org/10.2312/stag.20221263https://diglib.eg.org:443/handle/10.2312/stag20221263Computational methods to compute similarities between floor plans can help architects explore floor plans in large datasets to avoid duplication of designs and to search for existing plans that satisfy their needs. Recently, LayoutGMN [PLF*21] delivered state-of-the-art performance for computing similarity scores between floor plans. However, the high computational costs of LayoutGMN make it unsuitable for the aforementioned applications. In this paper, we significantly reduced the times needed to query results computed by LayoutGMN by projecting the floor plans into a common low-dimensional (e.g., three) data space. The projection is done by optimizing for coordinates of floor plans with Euclidean distances mimicking their similarity scores originally calculated by LayoutGMN. Quantitative and qualitative evaluations show that our results match the distributions of the original LayoutGMN similarity scores. User study shows that our similarity results largely match human expectations.Attribution 4.0 International LicenseCCS Concepts: Computing methodologies -> Modeling methodologiesComputing methodologiesModeling methodologies10.2312/stag.20221263115-1173 pages