一种服饰风格特征指导下的服装搭配学习模型  

Fashion Compatibility Learning Model Guided by Clothing Style Features

作　　者：刘锐 彭敦陆[1] LIU Rui;PENG Dun-lu(School of Optical Electrical and Computer Engineering,University of Shanghai Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《小型微型计算机系统》2022年第7期1378-1382,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61772342)资助.

摘　　要：时尚专家对于服饰搭配往往需要通过服饰的视觉属性(如颜色、图案及纹理等属性以及它们之间的组合)作为重要指导,进行有效地提取服饰视觉属性并用其改进传统的服饰搭配模型,对提升服饰搭配的有效性具有重要意义.本文在利用预训练的卷积神经网络中不同层次的卷积核来提取不同粒度的视觉属性(即视觉单词)的基础上,结合服饰的文字描述,采用多语言潜在迪利克雷分布模型进行多模态、无监督地挖掘出服饰风格特征.通过在双向长短时记忆模型中,加入了上述挖掘出的服饰风格特征作为训练指导,以此提升模型的计算效果.实验验证了本文提出的模型能够在服饰搭配的有效性上较其他方法有显著的提升.Fashion experts often need to use the visual attributes of clothing(such as color,pattern,texture and their combination)as important guidance to effectively extract the visual attributes of clothing and use them to improve the traditional fashion compatibility learning model,which is of great significance to improve the effectiveness of fashion compatibility learning.In this paper,the convolution kernel of different levels in the pre trained convolution neural network is used to extract the visual attributes(visual words)of different granularity.Combined with the text description of clothing,the multi-modal and unsupervised mining of clothing style features is carried out by using the polylingual Latent Dirichlet Allocation model.In order to improve the calculation effect of the model,the above-mentioned clothing style features are added into the Bidirectional Long Short-Term Memories as training guidance.Experimental results show that the proposed model can significantly improve the effectiveness of clothing matching compared with other methods.

关 键 词：视觉属性 风格特征 服装搭配 长短时记忆网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]