Automatic classification and segmentation of interior components from 3D scan models remain challenging, particularly for reconstructing semantically enriched interior representations. While various point cloud–based modeling approaches have been explored, the integration of deep learning–based component detection with NURBS-based reconstruction workflows in Grasshopper has received limited attention. This study presents an integrated plugin workflow that detects interior components in scanned environments and reconstructs them as parametric NURBS-based models.
This research introduces GeoVisionAI, a Grasshopper plugin that integrates deep learning–based object detection with parametric modeling. The framework employs a Mask R-CNN model trained on a custom dataset containing interior components, including walls, doors, windows and floors. The trained model performs detection and classification on data derived from 3D scans. Based on these results, the plugin automatically generates corresponding NURBS-based geometry within the Grasshopper environment, enabling the reconstruction of interior architectural elements.
The proposed framework successfully identifies interior components and converts detection outputs into parametric NURBS-based geometry. By integrating deep learning with parametric modeling workflows, GeoVisionAI enables the automated generation of semantically meaningful interior models. The framework supports cost-effective image-based 3D scanning using photogrammetry.
This study presents a novel framework that directly links deep learning–based component detection with automatic NURBS model generation within a parametric design environment. The approach contributes to digital reconstruction processes in architecture, engineering and construction.
