Research Information System
eCAADe2025, Ankara, Turkey, 01 September 2025, pp.1, (Full Text)
A critical factor for getting good results in machine learning (ML) applications is the presence of good quality and especially large datasets. ML is most especially applicable to the integration of the multi-dimensional aspects of architectural design. However, the current datasets are presented in various formats, which have inhibited the creation of large-scale datasets and pre-trained ML models. Furthermore, the current reliance on 2D architectural data is not enough to fulfill the data needs of architectural ML models, because design and the final outputs of architectural design are both 3D. BIM data could help to solve this issue since it contains rich geometric information as well as structural and spatial information. This paper is dedicated to the process of converting 3D BIM data for use as tensor data in ML applications. The proposed conversion method uses image stacking, which is used in medical imaging applications like MRI, CT scans and engineering applications. The paper gives a comprehensive account of how the geometric and class data from BIM is converted to image stack format at different levels of resolution, and how class data is encoded into pixels. This method is expected to help in the integration of 3D architectural representations into the machine learning environment in order to develop big data sets and pre-trained models. Also,BIM data is more diverse than other sources, it will help in identifying different relationships within the context of architectural machine learning