06.07.23 - From 3D Scans to 2D Floorplans: Innovations in BIM and Transformer Techniques

With the help of AI and machine learning technologies, Building Information Modeling (#bim ) can be used to analyze data from various sources such as sensors, cameras, and drones to provide insights into public administration or companies. Especially in the real estate industry, BIM can help reduce costs, improve efficiency, and increase productivity during construction or management. 
 
In Switzerland, Archilyse is a BIM company located in Zurich. Their product, 'Archilyse.one', is a cloud-based PropTech solution that offers AI-based API and SaaS services for the real estate industry. By analyzing the floor plan and building the BIM model semi-automatically, the customer can calculate rents and minimize vacancy more accurately and easier. Therefore, they can maximize the return on their portfolio. 
 
Besides floor plans or aerial images, 3D data like point clouds is also compatible with this kind of application. Recently, Yuanwen Yue et. ALL* from ETH Zürichhad been working on related projects and published their results on CVPR 2023. The paper proposes a new method for 2D floorplan reconstruction from 3D scans. Instead of using existing approaches that employ heuristically designed multi-stage pipelines, the authors formulate floor plan reconstruction as a single-stage task. 
 
To solve this problem, the authors develop a novel Transformer architecture that generates polygons of multiple rooms in parallel, in a holistic manner without hand-crafted intermediate stages. The method achieves a new state-of-the-art for two challenging datasets, Structured3D and SceneCAD, along with significantly faster inference than previous methods. Moreover, it can readily be extended to predict additional information such as semantic room types and architectural elements like doors and windows.
 
As the two projects mentioned above are open source, the STDL might have an interest in further developing the prototype for public administration. If you want to collaborate, do not hesitate to contact us.

Link to article : https://lnkd.in/ea3tcPbS

#BIM #realestate  #transformermodels  #pointcloud  #cvpr2023