摘要
服装图像实例分割是人工智能辅助服装设计中的关键环节,为实现对服装图像中细粒度属性的识别与定位,提出了基于深度学习的细粒度分割方法。该方法在原始Mask R-CNN的基础上,改进ResNet残差网络的结构,并在主干层网络中引入双向特征融合模块和双重注意力机制,从而提高模型的特征提取能力,在缩短信息路径的同时帮助网络模型关注更重要的区域。将该分割网络在iMaterialist Fashion数据集上进行验证与评价,结果表明该分割网络较原模型相比分割精度提高了约2.7%,该方法能够更加精准地进行细粒度实例分割,可为人工智能辅助服装设计的视觉系统研究提供新的思路。
Clothing images instance segmentation is a crucial section in artificial intelligence assisted fashion design.In order to realize the recognition and location of fine-grained attributes in clothing image,a fine-grained segmentation method based on deep learning was proposed in this paper.Based on the original Mask R-CNN,this method improved the structure of ResNet residual network,and introduced bidirectional feature fusion module and dual attention mechanism into the backbone layer network,so as to improve the feature extraction ability of the model and help the network model focus on more important areas while shortening the information path.The segmentation network was verified and evaluated on the iMaterialist Fashion dataset.The results show that the segmentation accuracy of the network is improved by about 2.7%compared with the original model.This method can segment fine-grained cases more accurately,and can provide a new idea for the research of visual system assisted by artificial intelligence in fashion design.
作者
王伟珍
赵汝嘉
WANG Weizhen;ZHAO Rujia(School of Fashion,Dalian Polytechnic University,Dalian,Liaoning 116034,China;Clothing Human Factors and Intelligent Design Research Center,Dalian Polytechnic University,Dalian,Liaoning 116034,China)
出处
《毛纺科技》
CAS
北大核心
2023年第6期88-94,共7页
Wool Textile Journal
基金
教育部社科规划基金项目(21YJAZH088)
教育部产学协同育人项目(220404211305120)
辽宁省教育厅基本科研重点攻关项目(LJKZZ20220069)
辽宁省教育厅教研项目(1010152)
中国纺织工业联合会教研项目(2021BKJGLX321)。
作者简介
第一作者:赵汝嘉,硕士生,主要研究方向为服装图像识别,E-mail:568597200@qq.com;通信作者:王伟珍,副教授,博士,主要研究方向为服装人因与智能设计,E-mail:wz-wang@foxmail.com。
