Abstract
Recent research into architectural form analysis using deep learning (DL) methods has shown potential to identify features from large collections of building data, shedding light into formal aspects of our built environment. As their application begins to impact architectural, urban, and policy design, it becomes important to critically engage with these techniques and their datasets.
For this work, we document the creation of a custom dataset of 331 3-D wooden churches located primarily within the Carpathian Mountain Regions of Ukraine and reveal morphological patterns that enhance existing scholarship on the subject. In particular, the complex rules that govern their vast and often entangled range of architectural styles and sub-styles. The dataset construction and analysis process resulted not merely from the implementation of advanced 3-D reconstruction and deep learning techniques, but also — and crucially — from subjective decisions, historical scholarship reviews, and expert and archival engagement. Documenting these, we illustrate how data collection, curation, and analysis are contingent upon social and technological factors, while identifying strengths and weaknesses, and opportunities for architectural-historical analyses to draw from historical traditions and state-of-the-art computational methods.
Indirectly, this paper also demonstrates an alternate means of architectural preservation through a combination of 3-D building reconstruction from sparse imagery and architectural style encoding using DL-methods. When considering the latest threats of war given Russia’s current invasion of Ukraine and targeted destruction of culturally significant buildings, computationally preserving both the churches themselves and their stylistic “essence” becomes increasingly important if we hope to fully document and protect these irreplaceable objects of Ukrainian architectural folk heritage for future generations.
