China (University of Hong Kong) HKU team wins a gold and a silver at the 2023 Conference on Computer Vision and Pattern Recognition
The AI algorithms for construction of 3D models of buildings, created by the research team led by Dr. Frank Xue at iLab of the Faculty of Architecture, beat competition from the world’s leading universities and won two awards – a gold and a silver – at the 2023 IEEE/ CVF Conference on Computer Vision and Pattern Recognition (CVPR).
The annual event, co-sponsored by the IEEE Computer Society Institute of Electrical and Electronics Engineers and the Computer Vision Foundation, is regarded as one of the most important conferences in its field. It was held in Vancouver of Canada this year.
The Conference involved a competition, called 3rd International Scan-to-BIM Challenge, where algorithms competed with each other in accuracy of reconstructing an existing building from the scanned data. Teams from around the world were given datasets (obtained by scanning laser scanning, or LiDAR) from real buildings in the United States. The datasets were extremely large, involving up to a billion data points. The team whose algorithm reconstructed the building most accurately won.
The HKU-iLab team claimed two awards in two separate categories. In the Scan-to-BIM 3D event (where a 3D model of a building is created from the data), the HKU scientists came first with their AI-based winning method called Space-voxel-guided Boundary Adaptation to Semantics Ensemble – Multi-start Optimization (SBASE-MO), beating the Tsinghua-CBIMS team from Tsinghua University’s National Research Center for Information Science and Technology (BNRist). A joint team from University of Kaiserslautern-Landau (RPTU) and German Research Center for AI (DFKI) came third.
In another category – Scan-to-BIM 2D (where a 3D model is converted to an architectural 2D drawing), the HKU team came second, with Tsinghua-CBIMS team in the first place. Here HKU-iLab team used another AI method called FLKPP++.
Building Information Modelling (BIM) is a technology that architects use to create a complete, and very detailed, digital model of a building. A BIM includes not only structures such as walls and airducts, but also furniture and appliances inside the building, as well as the entire set of information associated with the architectural project, for example the cost of each item and structure.
BIM is now a multi-billion-dollar global industry. Since 2020, the construction of all major (> 30 million HKD) public building projects in Hong Kong must incorporate BIM, and by 2026 BIM will be mandatory for major (> 300 million HKD) private building projects too.
To create a BIM of a building, the interior and the exterior are first scanned with sensors to record the position and the size of each structure and object. However, the problems arise when this data are converted into the model – the process is costly and laborious because of the large amount of data that needs to be crunched, and is also highly error prone.
For this reason, according to Dr Xue, there is now “an urgent need for Scan-to-BIM algorithms and software”. Not only such tools perform the entire process much faster, but they are also much more accurate than the current methods.
Dr. Frank Xue is an Assistant Professor at Department of Real Estate and Construction, HKU. He is Deputy Director of iLab and Vice Convener of Hong Kong-Zhuhai-Macao Research Station of the Key Scientific Research Base of Application of Spatial Information Technologies in Cultural Heritage Conservation of National Cultural Heritage Administration of China. He also serves as Vice-Chair of ACM-Hong Kong Chapter, committee of CGS-BIM Chapter, and committee of ASC-Smart Construction.
About iLab
iLab was established in 2016 under the Faculty of Architecture as an urban big data lab to take the opportunities and challenges as instigated by the global visions of Smart City and Industry 4.0. (FoA). It has made significant break-through in modernizing the construction industry in Hong Kong and beyond. The HKU iLab’s award-winning AI methods are a product of an ongoing ITF project granted by Hong Kong Innovation and Technology Commission (No.ITP/004/23LP), with supervision from Logistics and Supply Chain MultiTech R&D Centre (LSCM) and sponsorship from Paul Y. Engineering Group and UnoTech.