Intelligent generation model of packaging design based on dual-flow network and attention mechanism

Abstract

The existing design has obvious shortcomings in balancing local details with global styles, resulting in misalignment and poor coordination of design elements. Therefore, this study proposes an intelligent packaging generation model that integrates dual-stream network and attention mechanism, which extracts features in parallel by constructing content stream (CNN) and style stream (ViT) and combines cross-modal attention mechanism to achieve feature adaptive fusion. The experimental results show that the proposed model is significantly better than the comparative model in objective indexes such as FID (18.3) and PSNR (26.4) (for example, FID is 7.2 lower than the optimal baseline), the user satisfaction reaches 4.3/5.0, and it maintains good robustness in different noise environments (SSIM ≥0.79 under high noise). To sum up, this model effectively improves the automation level and generation quality of packaging design, and its innovative architecture provides an important practical reference for the application of AIGC in vertical fields.