| Title |
A Typological Analysis of Visual Datasets for Artificial Intelligence Research in Architecture |
| Authors |
허민지(Heo, Minji) ; 구형모(Gu, Hyeongmo) ; 추승연(Choo, Seungyeon) |
| DOI |
https://doi.org/10.5659/JAIK.2025.41.12.91 |
| Keywords |
Architectural datasets; Benchmark datasets; Artificial intelligence; Multimodal |
| Abstract |
As the use of artificial intelligence expands in the field of architecture, the importance of datasets has grown accordingly. However, prior
studies have mostly focused on reporting dataset construction, while relatively few have analyzed dataset characteristics. Since artificial
intelligence performance is closely linked to the scale, quality, and licensing of datasets, a comparative understanding of existing resources is
essential. This paper conducts a systematic review of major scholarly databases from 2005 to 2025 and proposes a two-level taxonomy for
datasets used in architectural artificial intelligence research. Representative datasets?such as floor plans, indoor 3D scans, aerial building
footprints, and BIM/CAD?are synthesized by type and compared in terms of modality, annotation, scale, and license. Furthermore, we
summarize dataset use cases along with citation and licensing information. Our analysis clarifies the strengths and limitations of each type
and provides a foundation for the development and application of robust datasets in future architectural artificial intelligence research. |