Distributed Computer Vision Pipeline for Fashion Generative Design: cGAN-Based Prototyping and Structural Equation Modeling of Designer-AI Creative Fit

Incorporating computer vision (CV) and generative artificial intelligence (AI) into fashion engineering, the AI-assisted prototyping has emerged as a fundamental technology in fashion engineering for enhancing the creative fit of fashion design. This study focuses on establishing a structural equation modeling (SEM) pipeline, constructed using both Python and SPSS AMOS, to explore the impact mechanism by which AI-assisted prototyping influences fashion generative design and aesthetic evaluation. In particular, the cGAN-based sketch-to-garment generation technique and the multi-task aesthetic evaluation architecture of ResNet-50 are integrated in the study. A total of 383 practitioners and researchers in the field of fashion design participated in the study. Five dimensions, namely, AI-assisted design inspiration, aesthetic preference, efficiency awareness, prototyping skills, and user experience, were developed to measure the degree of computational quantification in the AI-assisted fashion design process. A quantitative study combined with a computing-based pipeline was used in the research. Reliability and validity analyses, independent samples t-test, one-way analysis of variance, and SEM path analysis were conducted. The results showed that all the dimensions of the AI-assisted fashion design scale possessed high reliability, with all Cronbach's α coefficients above 0.8. The KMO value was 0.824, and the Bartlett sphericity test was significant (p<0.001), thereby validating the factor analysis. The cumulative variance explanation rate of the rotated factors reached 65.257%. A SEM pipeline combining computer technology was established and achieved excellent model fit (χ²/df=3.032, RMSEA=0.073). The cGAN-based prototyping engine in this study had an average time of 2.3 s for the inference of an image, while the ResNet-50-based multi-task aesthetic evaluation module achieved 87.3% accuracy in top-1 silhouette classification and 85.6% accuracy in texture classification. The findings suggest that there are selective paths through which AI generation affects creative design, and that including aesthetic evaluation in design process, optimizing efficiency, and enhancing creative fit are crucial.