基于生成对抗网络的中式婚服设计
Chinese Wedding Dress Design Based on Generative Adversarial Network
刘康 1马浩然 2邢乐1
作者信息
- 1. 江南大学数字科技与创意设计学院,江苏无锡 214400;江南大学江苏省非物质文化遗产研究基地,江苏 无锡 214122
- 2. 江南大学 设计学院,江苏 无锡 214122
- 折叠
摘要
为了解决传统中式婚服设计开发方法存在费时及效率低下的问题,提出将深度学习技术引入到中式婚服设计中,采用基于Pix2Pix算法模型的生成式设计方法,通过爬虫技术获取中式婚服图像数据,并对样本数据进行筛选以及轮廓特征、边缘特征和语义特征的标注,进而展开由单特征控制条件生成与特征联合控制条件生成两组实验.研究表明,联合控制条件生成的"递进式生成法"结合了生成对抗网络与条件图像生成方法的优势,服装特征标注被用作条件以增加服装图像生成过程的可控性,相较于"单特征控制条件生成"的细节调控能力更强,该结果可为中式婚服设计开发提供思路.
Abstract
In order to solve the time-consuming and low-efficiency problems in traditional Chinese wedding dress design and development methods,this study introduced deep learning technology into Chinese wedding dress design.It proposed a generative design method based on the Pix2Pix algorithm model.The Chinese wedding dress data was obtained through crawler technology.It annotated contour features,edge features and semantic features,and then launched two sets of experiments consisting of single feature control condition generation and feature joint control condition generation.The research shows that the"progressive generation method"that jointly controls conditional generation combines the advantages of generative adversarial networks and conditional image generation methods.Clothing feature annotations were used as conditions to enhance the controllability of the clothing image generation process.Compared to"single feature control condition generation",this approach offers superior detailed control capabilities,providing ideas for the design and development of Chinese wedding dress.
关键词
中式婚服/深度学习/Pix2Pix算法/控制条件生成Key words
Chinese wedding dress/deep learning/Pix2Pix algorithm/control condition generation引用本文复制引用
基金项目
江苏省哲学社会科学基金项目(21YSC009)
出版年
2024
NETL
NSTL
万方数据