Applications of Deep Learning in ArchitecturalFacade Parsing and 3D Reconstruction

Abstract The rapid advancements in deep learning have significantly transformed architectural facade parsing and 3D reconstruction,enabling precise and scalable analysis of urban environments. Traditional approaches, such as rule-based methods andhandcrafted feature extraction, often struggle with variations in architectural styles, occlusions, and perspective distortions.While convolutional neural networks (CNNs) have demonstrated considerable success in facade segmentation, they often lackstructural awareness, leading to inconsistent and fragmented parsing results. To address these challenges, we propose a novelHierarchical Structural Parsing Network (HSPN) that integrates multi-scale feature extraction, graph-based structural reasoning,and semantic refinement modules. Our method effectively captures both global architectural layout and fine-grained detailsby incorporating geometric constraints and adaptive structural consistency optimization (ASCO). Extensive experiments onbenchmark datasets demonstrate that our approach outperforms state-of-the-art methods in segmentation accuracy andarchitectural coherence. The proposed framework provides a robust solution for facade parsing and has potential applicationsin automated city modeling, digital twin construction, and immersive virtual environments.