Building façade datasets for analyzing building characteristics using deep learning

Area of Science:

• Computer Vision
• Machine Learning
• Architectural Engineering

Background:

• Building characteristics are crucial for construction management and architectural design.
• Static Street View Images (SSVIs) offer a non-invasive method for analyzing building attributes using deep learning.
• A significant gap exists in publicly available labeled datasets for training deep learning models on building facade imagery.

Purpose of the Study:

• To construct a novel, labeled dataset of building facade images.
• To facilitate the training of deep learning models for automated building characteristic analysis.
• To support four specific tasks: story count classification, building typology classification, exterior cladding material classification, and usable SSVI classification.

Main Methods:

• Sourcing facade images from London and Scotland, UK.
• Expert annotation of images for specific building characteristics.
• Dataset construction tailored for deep learning model development.

Main Results:

• A comprehensive dataset of labeled building facade images has been created.
• The dataset is suitable for training deep learning models for various classification tasks.
• The raw data can be repurposed for additional analyses with further annotation.

Conclusions:

• The developed dataset addresses the need for labeled building facade data.
• This resource enables automated analysis of building characteristics from SSVIs.
• The dataset has broad potential for future research in urban informatics and architectural studies.