文献类型：期刊文献

中文题名：基于循环一致性对抗网络的类推式汽车前脸生成设计

英文题名：Research on Generative Design of Analogous Car Front Face Based on CycleGAN

作者：季媛玮[1];陈育苗[1];潘岳[2]

机构：[1]华东理工大学艺术设计与传媒学院,上海200237;[2]中国科学技术大学软件学院,合肥230022

年份：2022

卷号：38

期号：6

起止页码：148

中文期刊名：机械设计与研究

外文期刊名：Machine Design And Research

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD_E2021_2022】；

基金：国家自然科学基金资助项目(51905175);上海市浦江人才计划(2019PJC021);教育部产学合作协同育人计划(202101042012)项目;上海市2021年度“科技创新行动计划”软科学重点项目(21692196800);上海市教委-高校紧缺艺术人才创新工作室(20212679)。

语种：中文

中文关键词：类比推理;生成式对抗网络;汽车前脸设计;智能设计

外文关键词：analogical reasoning;generative adversarial networks;car front face design;intelligent design

摘要：类比推理是设计师在产品外形设计阶段最常用的设计方法之一,然而设计结果严重依赖设计师经验,具有不确定性与随机性。为了辅助设计师进行创意生成,优化设计流程,提出一种基于类比推理的设计生成方法,结合循环一致性对抗网络(CycleGAN)以及相关图像处理技术,实现产品外形类比推理过程的智能化。首先,通过爬虫框架、主体目标检测与计算机视觉库(OpenCV)等处理汽车前脸轮廓,构建设计对象数据集。其次,收集图像构建启发源数据集。然后,使用CycleGAN对启发源数据集和设计对象数据集中的图像进行训练。最后,进行设计评价并整合训练数据,设计并实现原型系统推理者(Analogist),实现类比推理方案的智能生成。本项研究以猫的正面轮廓图像作为汽车前脸轮廓图像的类比推理对象,通过Analogist可以生成新的具有显著类比推理特征的汽车前脸轮廓图像。结果表明,该方法可以有效便捷地辅助设计师进行产品外形的类比推理设计,在一定程度上提升了设计效率,并为智能设计领域的发展提供了理论依据和实践基础。
Analogical reasoning is one of the most commonly used design methods in the product shape design stage.However,the design results rely heavily on the designer’s experience,which is uncertain and random.In order to assist designers in generating ideas as well as optimizing the design process,a design generation method based on analogical reasoning is proposed.Combined with cycle-consistent adversarial networks(CycleGAN)model and series of image processing methods,an intelligent analogical reasoning process of product shapes is achieved.Firstly,the design object dataset is constructed based on the Scrapy framework,subject target detection and computer vision library(OpenCV).Secondly,images are collected to build the inspiration dataset.Then,the two datasets are trained in the CycleGAN model.Finally,with the results evaluated and the training data integrated,the prototype system Analogist is designed and implemented.The intelligent generation of analogical reasoning is realized.The front view of the cat is considered as the analogical reasoning object of that of the car.This method shows that Analogist can generate new front images with analogical reasoning feature.The results demonstrate that this method can assist designers in analogical reasoning of product shape design and improve the design efficiency.This study provides the theoretical evidence and practical basis for intelligent design.