基于双鉴别器GAN的包装类产品外观设计法

针对传统包装类产品外观设计工作中设计周期较长、设计难度较大以及设计成本相对较高等问题,将双鉴别器GAN应用于包装类产品外观设计工作中。在BicycleGAN的基础上,添加了一个局部鉴别器,并重新构建了相应的损失函数以及目标函数。实验时,输入对象为 "边缘图像+原图"的图像对,输出对象为真实性概率排前10的图像。从多样性、PSNR值以及SSIM值3个方面对实验结果进行评估,结果表明:双鉴别器GAN生成的图像不仅拥有较好的多样性,其图像细节质量也有所提高。因此,将双鉴别器GAN应用于包装类产品的外观设计工作是可行的,一方面能够为设计师提供设计灵感,另一方面也节省了人力与物力成本,提高了工作效率。

Appearance Design Method of Packaging Product Based on Dual Discriminator GAN

To solve the problems of long design cycle, great design difficulty and relatively high design cost in the traditional packaging product design, the dual discriminator GAN was applied to the packaging product design. On the basis of BicycleGAN, a local discriminator was added, and the corresponding loss function and objective function were reconstructed. In the experiment, the input object was the image pair of "edge image + original image", and the output object was the image with the top 10 authenticity probability. The experimental results were evaluated from three aspects: diversity, PSNR value and SSIM value. The results showed that the image generated by dual discriminator Gan not only had better diversity, but also improved the image detail quality. Therefore, it was feasible to apply the dual discriminator Gan to the appearance design of packaging products. On one hand, it could provide designers with design inspiration, on the other hand, it could save human and material resources, and improve working efficiency.