A Foundation Model Approach: Generating Point Clouds for Scan-to-BIM Training and BIM Model Generation

Abstract: Modern scan-to-BIM approaches often use machine learning (ML) models to reconstruct BIM models from point clouds. Their performance relies heavily on a large amount of training data. Generating this data requires a significant amount of manual effort for outlier elimination and labelling. The approach uses a foundation model (FM) for the generation of point cloud training data. The presented generator enables the training of AI algorithms for the scan-to-BIM method without additional hardware. It also allows future researchers to generate data for a scan-to-BIM workflow.