Abstract: To solve the problem of information loss in the design process due to user's perceptual needs,a product form design method for cross modal data interaction was proposed.The relevant design elements were selected based on users' cognitive and emotional needs for product form and style.Based on the methods such as topic crawler and QFD,a sufficiently large "text image" product form dataset was constructed.Then,StackGAN was used to convert users' emotional needs to product form design solutions.The probability of target category was obtained by YOLO,and several design schemes that have a high degree of matching with users' emotional imagery was selected.In the proposed method,the user demand information with text as the main carrier was utilized as the main input to obtain product form design solutions.The cross modal imitation and reproduction of user needs from text to image was achieved with an end-to-end form,reducing the subjective understanding bias of designers towards user demand information.The feasibility of this method was verified using a household humidifier as an example.

Key words: product form, StackGAN, text to image, cross-modal, YOLO v4, generative design

摘要： 为解决用户感性需求在设计流程中的信息损耗问题,提出一种跨模态数据交互的产品形态设计方法。首先根据用户对产品的形态风格认知信息和感性需求信息筛选相关设计要素,基于主题爬虫、质量功能配置（QFD）,等方法,构建足够量级的“文本-图像”产品形态数据集,然后利用StackGAN实现用户感性需求到产品形态设计方案的转换,最后根据YOLO得到的目标类别概率,筛选与用户感性意象匹配程度较高的若干设计方案。本文提出的产品形态设计方法利用以文本为主导载体的用户需求信息为主要输入获取产品形态设计方案,以端到端的形式,实现用户需求从文到图的跨模态模仿与再现,减少设计师对于用户需求信息的主观理解偏差,并以家用加湿器为例验证其可行性。

关键词: 产品形态, StackGAN, 文本生成图像, 跨模态, YOLO v4, 生成式设计