Abstract This study presents an intelligent personalized garment customization system that integrates deep learning methodologies. The system employs a microservices architecture to unify four core modules: body measurement data extraction, style preference learning, virtual try-on visualization, and design recommendation generation. We propose a novel CNN-Transformer-GAN architecture, specifically tailored for personalized garment design tasks, achieving exceptional accuracy. Experimental results demonstrate that the system attains a mean absolute error (MAE) of 0.38 cm in body measurement, an accuracy of 87.4% in style matching, and a response time of 285 ms. Compared to existing approaches, the proposed system improves measurement accuracy by 38.7% and delivers visualization quality comparable to metaverse-based systems. To evaluate user experience, we conducted two complementary studies: (1) a controlled single-blind user study with 120 participants, which yielded satisfaction scores between 4.42 and 4.65 across recommendation accuracy, interface usability, and visualization quality; and (2) a large-scale deployment test involving 250 users, which reported an average overall satisfaction score of 4.55 out of 5.0. This research helps bridge the gap between artificial intelligence and personalized fashion design, advancing resource-efficient customization and better alignment with consumer needs in the apparel industry. By integrating state-of-the-art deep learning techniques with responsive user preference modeling, the system offers an innovative solution for intelligent garment customization.
Yeyue Lu (Wed,) studied this question.