Verification of AI Algorithms Assisted by Deep Learning in Automated Generation of 3D Models for Art and Design

As the core carrier of Shanghai’s regional culture, the unique layout of "Shikumen+Courtyard" and elements such as exposed brick wall in Shikumen architecture have both artistic value and historical significance. Currently, efficient 3D model generation methods are urgently needed in digital protection and innovative design. Problem: Traditional Shikumen 3D modeling relies on manual surveying and modeling, which has problems such as low efficiency, large deviation in detail restoration, and slow design innovation iteration, making it difficult to meet the needs of batch digitization and personalized design. The structure and content of this paper: this paper first constructs a dataset containing 12 core elements of Shikumen architecture, designs a four stage technical scheme of "feature extraction style transfer model generation detail optimization" based on the fusion algorithm of improved Generative Adversarial Network (GAN) and Convolutional Neural Network (CNN), to achieve automated generation of Shikumen 3D model; meanwhile, this paper introduces attention mechanism to enhance the restoration accuracy of characteristic elements such as door lintels and tiger windows. Experimental investigation results: in the adaptability test of multi style stone gate generation, the algorithm achieved success rates of 94.1%, 95.3%, and 91.8% for early, late, and improved combination of Chinese and Western stone gate generation, respectively, and it can automatically adjust the combination of architectural elements according to style keywords.