基于深度学习的汽车造型设计工具研究  

Automotive Modeling Design Tool Based on Deep Learning

作　　者：夏进军[1] 周方舟 樊真成 黄宝钢 刘礼 XIA Jin-jun;ZHOU Fang-zhou;FAN Zhen-cheng;HUANG Bao-gang;LIU Li(Chongqing University,Chongqing 400044,China)

机构地区：[1]重庆大学,重庆400044

出　　处：《包装工程》2021年第18期42-49,共8页Packaging Engineering

基　　金：重庆市社科规划项目(2018YBYS162);用户体验质量驱动的重庆智能产业产品设计多觉表达方法研究。

摘　　要：目的研究基于深度学习的汽车造型设计工具。方法使用交叉分类法对数据集进行整理并标记,形成全标签数据库,结合基于StyleGAN2的汽车图像生成模型,形成基于深度学习的汽车造型设计工具,并通过对比实验检验AI设计工具的生成速率和质量优劣。结果通过AI生成汽车造型的速率是一般设计师的数千倍,在质量比分中,AI生成的质量高于一般设计师。结论基于深度学习的汽车造型设计工具可作为汽车造型设计的新工具。This paper studies the automotive modeling design tools based on deep learning.The cross-classification method is used to arrange and mark the data set to form a full-label database.In combination with styleGAN2-based automobile image generation model,a vehicle modeling design tool based on deep learning is formed,and the generation rate and quality of AI design tools are tested through comparative experiments.Through the AI design,the automobile modeling generation rate is thousands of times higher than that of ordinary designers,and in the quality score,the quality generated by AI is higher than that of ordinary designers.By partially and controllable design target modeling,it is proved that the automobile modeling design tool based on deep learning is a new tool,and the rate of generating images by the new tool is faster than that of the designer’s hand drawing.

关 键 词：汽车造型 深度学习 设计工具 StyleGAN2 标签体系 

分 类 号：TB472[一般工业技术—工业设计]