[Paper Review] SAGC-A68: a space access graph dataset for the classification of spaces and space elements in apartment buildings

Amir Ziaee, Georg Suter|arXiv (Cornell University)|Jul 10, 2023

3D Modeling in Geospatial ApplicationsEngineering3 citations

TL;DR

This paper introduces SAGC-A68, a novel graph-based dataset of 68 3D apartment building layouts, where space access graphs are automatically generated to enable Graph Deep Learning (GDL) for classifying 22 space functions and 6 space elements. The authors demonstrate its utility by training a Graph Attention Network (GAT) achieving strong classification performance, with code and data publicly released to support future research in automated building space analysis.

ABSTRACT

The analysis of building models for usable area, building safety, and energy use requires accurate classification data of spaces and space elements. To reduce input model preparation effort and errors, automated classification of spaces and space elements is desirable. A barrier hindering the utilization of Graph Deep Learning (GDL) methods to space function and space element classification is a lack of suitable datasets. To bridge this gap, we introduce a dataset, SAGC-A68, which comprises access graphs automatically generated from 68 digital 3D models of space layouts of apartment buildings. This graph-based dataset is well-suited for developing GDL models for space function and space element classification. To demonstrate the potential of the dataset, we employ it to train and evaluate a graph attention network (GAT) that predicts 22 space function and 6 space element classes. The dataset and code used in the experiment are available online. https://doi.org/10.5281/zenodo.7805872, https://github.com/A2Amir/SAGC-A68.

Motivation & Objective

• To address the lack of suitable datasets for training Graph Deep Learning (GDL) models in building space classification.
• To reduce manual effort and errors in classifying spaces and space elements in architectural models.
• To enable automated, data-driven classification of space functions and elements in apartment buildings using graph representations.
• To provide a reproducible benchmark dataset for evaluating GDL models in architectural space understanding.
• To support applications in building energy efficiency, safety analysis, and usable area assessment through accurate space classification.

Proposed method

• The authors automatically generate space access graphs from 68 digital 3D models of apartment building layouts.
• Each graph encodes spatial adjacency and connectivity between rooms and space elements, capturing functional and structural relationships.
• The dataset is constructed by extracting topological relationships between spaces, treating them as nodes and edges in a graph.
• A Graph Attention Network (GAT) is trained on the SAGC-A68 dataset to classify 22 space function classes and 6 space element classes.
• The model leverages attention mechanisms to weigh the importance of neighboring nodes in the graph during classification.
• The entire pipeline, including data preprocessing and model training, is implemented and released with the dataset for reproducibility.

Experimental results

Research questions

• RQ1 Can a graph-based representation effectively encode spatial relationships for classifying space functions and elements in apartment buildings?
• RQ2 How well can a Graph Attention Network (GAT) perform on a newly introduced, real-world architectural graph dataset like SAGC-A68?
• RQ3 To what extent does the graph structure of space access improve classification accuracy compared to alternative representations?
• RQ4 Can the proposed dataset serve as a reliable benchmark for future Graph Deep Learning models in architectural space analysis?
• RQ5 What is the generalization potential of GDL models trained on this dataset for real-world building analysis tasks?

Key findings

• The SAGC-A68 dataset successfully captures the spatial topology of 68 real apartment building layouts in a graph format suitable for GDL.
• The trained GAT model achieves strong performance in classifying 22 space function and 6 space element categories on the SAGC-A68 benchmark.
• The graph-based representation enables effective modeling of spatial connectivity and functional relationships between rooms.
• The dataset and code are publicly released, enabling reproducibility and fostering future research in automated space classification.
• The results demonstrate the feasibility and effectiveness of using GDL with access graphs for architectural space understanding in building information modeling (BIM).