Deep Learning Driven Method to Analyze Large Scale Dataset of Korean Apartment Unit Floor Plan (2020—2021)

(Published in Journal of the Korean Institute of Interior Design & CAADRIA 2021)

Kyung Hoon Hyun

Floor plan analysis; Design quantification; Spatial pattern analysis; Semantic fingerprint; Residential layout

  • Limitations of previous research from an insufficient number of collected floor plan datasets and analyzed features were overcame with reliable analysis.
  • A data-driven method was proposed to analyze apartments and to generate reliable results
  • 5,344 floor plans of the apartments were quantitatively investigated and the spatial pattern of Korean apartments was established.

This paper introduces a unique quantitative analysis method and results that are differentiated from those in existing studies. We analyzed five types of information in floor plan images: the silhouette, number of rooms, room area, and direct and indirect room connectivity. Furthermore, the analysis used a large-scale apartment unit dataset consisting of 33,892 units. We present convincing and objective spatial pattern analysis results of Korean apartments by quantitatively analyzing a large-scale dataset. It is expected that the analysis results will clarify the characteristics of the residential environment of Korean apartments. The results suggest that changes in lifestyles lead to the modularization of bedrooms, increased numbers of private bathrooms and balconies with corridors as junctions, and the diversification of room layouts.

  1. H. Maeng, K. H. Hyun., "A Deep Learning Driven Method to Analyze Large Scale Dataset of Korean Apartment Unit Floor Plan - Focused on the Korean Apartment from 1970 to 2020 -." Journal of the Korean Institute of Interior Design, vol. 30, no. 3, 30 June 2021, pp. 65–76.
  2. H. Maeng, K. H. Hyun., "Data-Driven Analysis of Spatial Patterns through Large-Scale Datasets of Building Floor Plan." CAADRIA 2021: Proceedings of the 26th Conference on Computer-Aided Architectural Design Research in Asia, 2021, pp. 301-310.

© Copyright 2023 by Hoyoung Maeng. All Rights Reserved.