In recent years, there has been a growing interest in automated indoor construction progress monitoring (ICPM) to maximize precision and reduce human intervention. Computer vision approaches, especially based on deep learning (DL) methods, have shown great potential in this task. However, training DL models require large-scale datasets, which are often costly and laborious to obtain, specifically for indoor construction environments. This study proposes an automated approach to generate real-world-like synthetic data of indoor construction by combining building information modeling (BIM) and a photo realistic graphics engine. The approach was validated by efficiently producing annotated synthetic datasets of mechanical, electrical, and plumbing components from various BIM models. A state-of-the art instance segmentation network was trained using those datasets alongside real manually annotated data and transfer learning methods to assess the results. Preliminary experiments using an on-site augmented reality device demonstrate the promising efficiency of DL for ICPM.