Pre-Emptive 3D Visualization of Architectural Blueprints using Deep Learning

The traditional way to visualize architectural blueprints relies on manual 3D modelling. This process requires significant expertise and efforts. In the field of architectural design, traditionally, the conversion from 2D blueprints to 3D models is a manual, labour intensive, time-consuming, and error-prone process. This paper proposes an AI-driven system that automates this conversion process using Deep Learning. The approach integrates Computer Vision and Generative Modelling methods to preprocess floor plans, extract the structural features, and reconstruct the 3D layouts. This paper dives deeper into using datasets such as Plan2Scene, Structured3d, etc., along with the generation of synthetic data to reduce modelling time significantly. The results are evaluated on parameters such as Intersection over Union (IoU), Structural Similarity Index (SSIM), Learned Perceptual Image Patch Similarity (LPIPS), and time efficiency. The goal was to demonstrate a significant reduction in the turnaround compared to manual CAD workflows. The proposed solution enhances client engagement while accelerating design cycles and immersive visualization in architecture.