The research centered on applying deep learning (DL)-based point cloud completion algorithms to overcome the issue of missing or obstructed data in MEP scans. To do this effectively, the team needed a large dataset of synthetic point clouds derived from realistic BIM models of MEP systems. These point clouds would include controlled occlusions, allowing them to evaluate the performance of multiple DL models and training strategies.
This required a reliable method to convert Revit BIM models into OBJ format, from which synthetic point clouds could be simulated using tools like CloudCompare or custom Python scripts.