CADSpotting accurately identifies and segments symbols in CAD drawings, facilitating tasks like 3D interior modeling. It uses dense point sampling within a unified point cloud model to learn robust primitive features, and integrates Sliding Window Aggregation for efficient panoptic symbol spotting in large-scale drawings. The resulting semantic information enables automated parametric reconstruction of architectural 3D interiors.
We introduce CADSpotting, an effective method for panoptic symbol spotting in large-scale architectural CAD drawings. Existing approaches often struggle with symbol diversity, scale variations, and overlapping elements in CAD designs, and typically rely on additional features (e.g., primitive types or graphical layers) to improve performance. CADSpotting overcomes these challenges by representing primitives through densely sampled points with only coordinate attributes, using a unified 3D point cloud model for robust feature learning. To enable accurate segmentation in large drawings, we further propose a novel Sliding Window Aggregation (SWA) technique that combines weighted voting and Non-Maximum Suppression (NMS). Moreover, we introduce LS-CAD, a new large-scale dataset comprising 45 finely annotated floorplans, each covering approximately 1,000 2, significantly larger than prior benchmarks. LS-CAD will be publicly released to support future research. Experiments on FloorPlanCAD and LS-CAD demonstrate that CADSpotting significantly outperforms existing methods. We also showcase its practical value by enabling automated parametric 3D interior reconstruction directly from raw CAD inputs.
Overview of the CADSpotting pipeline. (a) Given a CAD drawing as input, CADSpotting first densely samples points along CAD graphic primitives to construct a point cloud representation, where each point is defined by its spatial coordinates.(b) Next, PTv3 extracts robust geometric features from the sampled point cloud.(c) Finally, after aggregating primitive-level features via mixed pooling, a streamlined Transformer decoder is employed 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 comprising 45 floorplans from expansive buildings, such as campuses and office complexes. Each drawing covers at least 1,000 square meters, with the number of primitives rangng from approximately 2,900 to over 10,000. The dataset features a rich variety of complex symbols, including doors, windows, walls, elevators, and parking spaces. All floorplans are annotated with fine-grained labels consistent with the FloorPlanCAD (Fan et al. 2021) standard. The LS-CAD dataset will be publicly released to support future research in panoptic symbol spotting for large-scale CAD drawings.
CADSpotting produces accurate panoptic segmentation results that serve as the foundation for our parametric 3D interior modeling pipeline. Using instance-level segmentation and primitive positions, we extract spatial parameters for key architectural elements, including walls, doors, and windows. For doors, we compute positions, orientations, and pivot points by analyzing geometric relationships between arcs and lines within each segmented instance. Window positions are directly derived from their corresponding instances. Wall modeling involves a more specialized process: we merge endpoints of adjacent lines within wall instances to form closed polygons, which are rasterized into binary masks. Final wall contours are then extracted using the method from (Suzuki and be 1985). These spatial parameters are fed into a parametric reconstruction pipeline implemented in Blender. Predefined 3D assets for walls, doors, and windows.
BibTex Code Here