Our CADSpotting accurately identifies and segments symbols in CAD drawings, thereby enhancing the efficiency of tasks such as 3D interior modeling. The approach employs a dense point sampling method within a unified point cloud processing model to improve primitive feature representation. Additionally, a Sliding Window Aggregation technique is incorporated during inference to facilitate panoptic symbol spotting in large-scale CAD drawings.
We introduce CADSpotting, an efficient method for panoptic symbol spotting in large-scale architectural CAD drawings. Existing approaches struggle with the diversity of symbols, scale variations, and overlapping elements in CAD designs. CADSpotting overcomes these challenges by representing each primitive with densely sampled points, described by essential attributes like coordinates and colors. Building upon a unified 3D point cloud model for joint semantic, instance, and panoptic segmentation, CADSpotting learns robust feature representations. To enable accurate segmentation in large, complex drawings, we further propose a novel Sliding Window Aggregation (SWA) technique, combining weighted voting and Non-Maximum Suppression (NMS). Moreover, we introduce a large-scale CAD dataset named LS-CAD to support our experiments. Each floorplan in LS-CAD has an average coverage of 1,000 m2 (versus 100m2 in the existing dataset), providing a valuable benchmark for symbol spotting research. Experimental results on FloorPlanCAD and LS-CAD datasets demonstrate that CADSpotting outperforms existing methods, showcasing its robustness and scalability for real-world CAD applications.
Overview of the CADSpotting method: Given a CAD drawing as input, CADSpotting first densely samples points along CAD graphic primitives to build a comprehensive point cloud representation, with each point defined by its coordinates and color. Next, Point Transformer V3 (PTv3) serves as the backbone, extracting robust features from the sampled points. We then apply mixed pooling to obtain the primitive-level features. Finally, a streamlined Transformer decoder is used for efficient panoptic symbol spotting.
Qualitative comparison of panoptic symbol spotting. Our method accurately detects symbol instances, even in situations where symbols are overlapped by other elements.
We introduce LS-CAD, a new large-scale CAD dataset consisting of 50 floorplans from extensive buildings, such as campuses and office buildings. Each CAD drawing in LS-CAD covers an area of at least 1,000 square meters, with the number of primitives ranging from approximately 2,900 to over ten thousand. Additionally, the floorplans in LS-CAD include a wide variety of complex symbols, such as doors, windows, walls, elevators and parking spaces. We provide fine-grained annotations consistent with the standards of the FloorPlanCAD~\cite{fan2021floorplancad} dataset. A portion of LS-CAD will be publicly available under a license waiver to support further research in panoptic symbol spotting for large-scale CAD drawings.
We also demonstrate the effectiveness of CADSpotting by automating the generation of 3D interior models through parametric modeling. Using precise instance-level data obtained from CADSpotting, we compute spatial parameters for architectural elements, including wall contours, door and window positions, door orientations, and pivot points. With these spatial parameters clearly defined, we leverage parametric reconstruction techniques within Blender to efficiently build 3D interior models directly from raw CAD data. This approach significantly streamlines the modeling process, enhancing productivity and accuracy. To the best of our knowledge, our proposed pipeline is the first comprehensive solution that seamlessly integrates CAD segmentation with automated 3D interior model reconstruction.
BibTex Code Here