Enhancing Deep Learning-based BIM Element Classification via Data Augmentation and Semantic Segmentation

A critical aspect of BIM is the capability to embody semantic information about its element constituents. To be interoperable, such information needs to conform to the Industry Foundation Classes (IFC) standards and protocols. Artificial intelligence approaches have been explored as a way to verify the semantic integrity of BIM to IFC mappings by learning the geometric features of individual BIM elements. The authors through previous studies also investigated the use of geometric deep learning to automatically classify individual BIM element classes. However, such efforts were limited in the number of training data and restricted to subtypes of BIM elements. This study has significantly expanded the training set, to include a total of 46,746 elements representing 13 types of BIM elements. The magnitude of the data set is considered to be the first of this kind. Furthermore, Conditional Random Fields as Recurrent Neural Networks (CRF-RNN), a deep learning algorithm for semantic segmentation, was deployed to enhance the quality of individual input images. Deploying the dataset and segmentation improved the performance of previous model (Multi-View CNN) by 4.37% and achieve an overall performance of 95.38%.