In recent years, the integration of computer vision with culturally informed industrial design has emerged as a compelling direction for intelligent 3D modeling, particularly in scenarios requiring spatial adaptability and semantic coherence. Traditional three-dimensional design methodologies, often based on rule-driven CAD systems or purely geometric logic, fall short when faced with complex, context-rich environments. These approaches tend to overlook critical observational diversity and cultural nuance, resulting in models that struggle with generalization, multimodal integration, and structural fidelity in real-world design tasks. Addressing these limitations, this study presents a unified vision-driven framework for three-dimensional product structure optimization that effectively merges multimodal visual data with traditional aesthetic design principles. The proposed framework is composed of two tightly integrated components: the Cross-Modal Vision Aggregator (CMVA) and the Layout-Guided Iterative Refinement Mechanism (LGIRM). CMVA fuses high-density aerial imagery with sparse ground-level views using convolutional backbones, adaptive kernel interpolation, and multi-scale feature embedding to produce spatially aligned, semantically rich representations of target regions. Simultaneously, LGIRM ensures structural consistency and semantic flow by introducing layout-sensitive regularization and iterative refinement procedures during model training. Together, these modules facilitate a design process that is both responsive to real-world variability and capable of preserving cultural and structural integrity. Experimental validations demonstrate the framework’s superior performance in generating context-aware, geometrically coherent designs compared to conventional baselines. This work lays a foundation for advancing intelligent, vision-based design systems that bridge technical precision with cultural sensitivity in industrial applications.
Libo Lu (Wed,) studied this question.