基于生成对抗网络(Pix2pix)的家具设计草图渲染  

Furniture Design Sketch Rendering Based on Generative Adversarial Network(Pix2pix)

作　　者：朱文霜[1] 王禹钧 郑文俊 Zhu Wenshuang;Wang Yujun;Zheng Wenjun

机构地区：[1]桂林理工大学艺术学院,广西桂林541006

出　　处：《家具与室内装饰》2023年第10期117-121,共5页Furniture & Interior Design

基　　金：国家自然科学基金项目(52368005);广西高等教育本科教学改革工程项目(2022JGA208)。

摘　　要：家具设计师工作初期通常会借助手绘线稿来描绘创意与灵感,但手绘线稿往往难以快速深化效果。为改善该状态,故尝试运用生成对抗网络(Generative Adversarial Network, GAN)门类中的深度学习图像转译模型(Pix2pix)对手绘线稿进行快速渲染。先以样本采集法创建家具数据集,据此对Pix2pix进行训练;再建立交互界面来满足用户操作需求;最后通过两组测试验证其可行性。通过对Pix2pix生成的图像效果与使用者的反馈分析可知,Pix2pix能有效辅助家具设计师进行草图渲染,同时也能帮助非专业设计人员呈现自身创意。这既为家具设计效果表现探索出一条数字化路径,也是一次设计学和计算机科学与技术的深度融合研究。Furniture designers often use hand-drawn sketches to depict their creativity and inspiration in the early stages of their work,but hand-drawn sketches are often difficult to deepen quickly.To improve this situation,they attempted to use the deep learning image translation model(Pix2pix)in the Generative Adversarial Network(GAN)category to quickly render hand-drawn sketches.First,a furniture dataset was created using sample collection methods,and Pix2pix was trained based on this dataset.Then,an interactive interface was established to meet user operation requirements.Finally,the feasibility was verified through two sets of tests.Through the analysis of the image effects generated by Pix2pix and user feedback,it was found that Pix2pix can effectively assist furniture designers in sketch rendering and help non-professional designers present their own creativity.This not only explores a digital path for furniture design performance but also represents a deep integration of design,computer science,and technology research.

关 键 词：生成对抗网络 图像转译 深度学习 家具设计 Pix2pix 

分 类 号：TS664.1[轻工技术与工程]