基于深度学习的男装款式图制版要素识别  

Identification of pattern elements in men's fashion design sketches based on deep learning

作　　者：胡志霖 陈敏之[2] HU Zhilin;CHEN Minzhi(School of Fashion Design and Engineering,Zhejiang Sci-Tech University,Hangzhou 311100,China;School of International Education,Zhejiang Sci-Tech University,Hangzhou 311100,China)

机构地区：[1]浙江理工大学服装学院,浙江杭州311100 [2]浙江理工大学国际时装技术学院,浙江杭州311100

出　　处：《染整技术》2024年第9期66-72,115,共8页Textile Dyeing and Finishing Journal

摘　　要：探讨在传统服装制版领域中如何应用深度学习技术,以替代传统制版师在款式图中进行尺寸评价关键点和部件类别信息提取的人工思维过程,从而提高服装结构设计的效率和质量。首先,对经典服装款式图的构成要素进行了分析,以及探讨了制版师在款式图分析中提取关键信息的思维模式。其次,介绍了基于Hourglass和YOLOv5款式图识别模型的设计和实现过程。对包含15个服装尺寸测量关键点的数据集和包含15种常见男装部件的数据集进行了预处理、标注和数据增强等,共涵盖2000张男装款式图。对这两个数据集进行训练,迭代了多个模型,并对模型的性能指标进行了测试、评估和分析。深度学习技术成功实现了对款式图中关键信息的自动提取,有效替代了传统制版师的人工判读过程,从而提高了设计效率和质量。未来的研究方向包括扩大数据集规模、丰富标签类别、探索更复杂的模型结构,并考虑其他领域的应用,如女装、童装款式图识别等。The application of deep learning technology in the traditional field of fashion pattern making is explored,aiming to replace the manual cognitive process of pattern makers in extracting key dimension points and component category information from design sketches.This approach seeks to improve the efficiency and quality of fashion structural design.Firstly,the constituent elements of classic fashion design sketches are analyzed,and the cognitive patterns of pattern makers are investigated when extracting key information from these sketches.Secondly,the design and implementation process of a design sketch recognition model based on Hourglass and YOLOv5 is introduced.A dataset comprising 15 key measurement points for garments and another dataset containing 15 common men's clothing components are preprocessed,annotated,and augmented,covering a total of 2000 men's fashion design sketches.These datasets are used to train multiple models iteratively,and the performance metrics of these models are tested,evaluated,and analyzed.Deep learning technology successfully achieved the automatic extraction of key information from design sketches,effectively replacing the traditional manual interpretation process of pattern makers,thereby improving design efficiency and quality.Future research directions include expanding the dataset scale,enriching label categories,exploring more complex model structures,and considering applications in other fields such as women's and children's fashion design sketch recognition.

关 键 词：深度学习 款式图信息识别 制版流程优化 HOURGLASS YOLOv5 

分 类 号：TS941.7[轻工技术与工程—服装设计与工程]