NovalisArch+: Efficient hybrid procedural-AIGC for high-fidelity 3D city generation

Abstract

3D city modeling is a key area of computer vision and powers smart city applications. However, existing methods, such as procedural modeling and deep learning-based approaches, either require heavy manual intervention in asset creation or suffer from ”blind zone effects” caused by viewpoint limitations, thereby limiting the cost-effectiveness and interactivity of urban scene generation. To address these challenges, we introduce NovalisArch+, a lightweight 3D reconstruction framework that synergistically integrates procedural content generation (PCG) and AI-generated content (AIGC) to enable asset-efficient urban modeling. By encoding region-specific aesthetic rules for Chinese cities and unifying GIS data analysis, rule-driven architectural generation, and AIGC-driven physically based rendering (PBR) material synthesis, NovalisArch+ establishes a scalable and efficient pipeline for constructing semantically coherent, stylistically consistent 3D urban environments.