Exploring the Automated Process of Visual Creative Design Assisted by Generative Adversarial Networks (GANs)

The application of generative adversarial networks (GAN) in the field of visual creative design is gradually becoming a research hotspot. The current design industry faces problems such as low efficiency of creative generation, high labor costs, and single design style. This paper will explore how GAN can help the automation process of visual creative design by building a design generation framework based on GAN. First, a large-scale design dataset is collected and preprocessed, including image classification, annotation and normalization. Secondly, the generator and discriminator networks are constructed. The generator adopts the U-Net structure to capture detailed information, and the discriminator adopts the PatchGAN structure to improve local authenticity. Then, the Adam optimizer is used to perform adversarial training on the generator and the discriminator, and the model is optimized by minimizing the generation loss of the generator and maximizing the discrimination loss of the discriminator. Then, the conditional GAN (cGAN) technology is introduced to control the style and content of the generated samples by inputting design style labels or sketches as conditional information. The framework has significantly improved the diversity and quality of generated design samples. The SSIM value of the generated design samples is as high as 0.98, and the design generation time is as low as 0.3 min. GAN technology can effectively improve the automation level of visual creative design and provide an efficient and low-cost solution for the design industry.