Intelligent generation method of fashion design images based on CycleGAN

Abstract

Fashion design is a comprehensive art that combines creativity, aesthetics, technology, and other aspects. By designing and combining elements such as clothing styles, colors, and image styles, fashion works with unique styles and practical functions are created. However, the existing clothing design methods all have defects such as poor image accuracy and long generation cycle. Therefore, this study optimizes the Cycle Consistency-Generative Adversarial Network by introducing Least Squares and Efficient Channel Attention Network, and constructs a brand-new intelligent fashion design image generation model based on this. The results show that this model achieves an accuracy of 97.1%, an average success rate of 96.8%, and a 37 dB image signal-to-noise ratio, significantly outperforming the comparison models. Meanwhile, when generating a substantial volume of clothing designs, the model maintained a peak response time of merely 229 ms and a minimal loss rate of 0.17%. And it achieved a score of 97.3 in a questionnaire based on the esthetic ratings of 200 users. In conclusion, the proposed model not only generates higher-quality, more detailed images but also ensures high precision and success rates. The clothing designs are widely accepted by the public, meet modern esthetic standards, and effectively address the challenges of existing fashion design methods, and is conducive to designing higher-quality fashion garments.