Clothing Customization Design based on Style Generative Adversarial Network with Cross-Model Recommendation

In recent years, fashion technology played a vital role in developing the modern world by integrating style transfer technology for personalized clothing and customized design. However, this transfer technology was limited with shorter average runtime which leads to low processing rate of higher resolution images. To overcome this challenge, the proposed Style Generative Adversarial Network with Cross-Model Recommendation (StyleGAN-CMR) utilizes Content-Style Image (CSI) dataset, which comprises content, style and artifact images. Further, collected data pre-processed with min-max normalization techniques by transforming image pixel values into standard range. Then, Visual Geometry Group 16 (VGG16) is used to extract features from normalized images and fed into StyleGAN model. This model consists self-attention mechanism-based generator to generate clothing design image. By analyzing generated images helps to recommend clothing designs based on factors such as color palette, style and patterns to create personalized list of clothing designs recommendations by using CMR, realize the customized design of clothing to meet the personalized clothing demands. The proposed StyleGAN - CMR model obtained better results when compared with exiting CycleGAN in terms of Mean Squared Error (MSE) (0.01), Normalized Mean Absolute Distance (NMAD) (0.01) and average runtime (0.03) respectively.