Application of Cycle Generative Adversarial Networks for Unpaired Image Style Transfer in Product Packaging Design

Abstract

The rapid advancement and extensive application of AI technology in the past few years have also spurred rapid evolution in deep learning. Image recoloring, detection, recognition, and creative style transfer are some of the most common applications of deep learning technology. Visual art styles are also frequently transferred using methods based on deep learning. The goal of this paper is to develop an algorithm for rapid visual art style migration based on building confrontation networks. The CelebA-HQ celebrity faces dataset, collected from multiple sources of artistic and product design images, was used to assess the effectiveness of style transfer. The preprocessing stage involves resizing all images to a fixed resolution and Min-max normalizing, ensuring consistent input and stable GAN training for effective visual style migration. A feature extraction technique, such as Histogram of Oriented Gradients (HOG), is employed to evaluate structural consistency before and after style migration. This paper introduces a novel method for automatically generating image-quality art using the Cycle Generative Adversarial Network (CycleGAN) and a migration algorithm. The CycleGAN produced high reconstruction quality with a PSNR of 23.84 and strong structural consistency with an SSIM of 0.81, confirming its effectiveness in artistic style migration for packaging design. To enhance the effect of image creative style migrations, the image is identified using a multiscale discriminator. The results of the trial show that this strategy is very valuable and applicable for advertising and use.