基于DCGAN算法的服装效果图生成方法  

Design method of garment effect drawing based on DCGAN algorithm

作　　者：郭宇轩 孙林[1] GUO Yuxuan;SUN Lin(School of Fashion,Dalian Polytechnic University,Dalian,Liaoning 116034,China)

机构地区：[1]大连工业大学服装学院,辽宁大连116034

出　　处：《毛纺科技》2024年第2期114-120,共7页Wool Textile Journal

基　　金：辽宁省“兴辽英才计划”项目(XLYC2210025);辽宁省教育厅科研项目(LJKFR20220220)。

摘　　要：为了提高服装设计效率,适应时尚产品迭代加速的趋势,提出一种基于深度卷积对抗网络(DCGAN)的服装效果图生成方法。搭建适用于服装效果图生成任务的DCGAN模型,制作服装秀场数据集进行模型训练并生成服装效果图,设计师主观筛选具有设计参考价值的生成服装效果图,计算有效生成图像比例,评估该模型性能和生成图像质量,通过人机交互的方式优化部分生成图像并形成最终设计方案。结果表明:优化后的DCGAN模型可以快速提取流行趋势生成创意设计方案,辅助设计师高效完成设计效果表达,为服装设计的智能化提供有效途径和方法参考。In order to improve the efficiency of clothing design and adapt to the trend of accelerating fashion product iteration,a method of generating fashion illustration based on deep convolutional Adversarial network(DCGAN)was proposed.By building a DCGAN model suitable for the task of fashion illustration generating,making a clothing show data set for model training and fashion illustration generating,the designer subjectively selects the generated fashion illustration with design reference value,calculates the proportion of effectively generated images,and evaluates the performance of the model and the quality of generated images.Through human-computer interaction,the optimized part generates images and forms the final design scheme.The results show that the optimized DCGAN model can quickly extract fashion trends and generate creative design schemes,assist designers to efficiently complete the expression of design effects,and provide effective ways and methods for the intelligent fashion design.

关 键 词：卷积神经网络 DCGAN 服装效果图 交互设计 深度学习 

分 类 号：TS941.26[轻工技术与工程—服装设计与工程]