文献类型：会议论文

英文题名：Research on Generative Design of Car Side Colour Rendering Based on Generative Adversarial Networks

作者：Ji, Yuanwei[1];Chen, Yumiao[1]

机构：[1]East China Univ Sci & Technol, Sch Art Design & Media, 130 Meilong Rd, Shanghai, Peoples R China

会议论文集：24th International Conference on Human-Computer Interaction, HCI International 2022 (HCII)

会议日期：JUN 26-JUL 01, 2022

会议地点：Goteborg, SWEDEN

语种：英文

外文关键词：Generative adversarial networks; Car colour rendering; MUNIT

摘要：The traditional artificial color rendering process often relies on the limited personal design experience of the designer. Therefore, the design results are often uncertain. In order to help designers get inspiration for color rendering, a multimodal generation method for car side color rendering schemes was proposed through the multimodal unsupervised image translation framework MUNIT based on generative adversarial network. Firstly, based on the image crawling technology and image batch collection tools, the car colour rendering inspiration dataset consisting of hand-drawn colour pictures of the car side was constructed. After that, the car side hand-drawn images were processed through the image styling processing technology and deep learning pre-trained models to construct the car color rendering design object dataset. Next, the car color rendering generation experiment was conducted through MUNIT framework. Then, the generated images were evaluated in quantitative and qualitative methods to select the best iterative model. Finally, with the experimental data integrated, the intelligent generative design system Analogist for car color rendering was designed. The results of the research showed that the method we proposed can realize the multimodal generation of color rendering schemes of line drawings of car side through an image-to-image inspired approach. It can be concluded that Analogist can assist designers in stimulating design inspiration and improving design efficiency in colour rendering of car styling.