计算机与数字工程2025，Vol.53Issue(3)：829-834,6.DOI:10.3969/j.issn.1672-9722.2025.03.036
基于属性解纠缠表示的交互式服装图像检索
Interactive Clothing Retrieval Based on Attribute Disentangled Representations
摘要
Abstract
Interactive clothing retrieval can replace the specific attributes in the query image with the target attributes to up-date the retrieval search results.However,there are semantically entangled in the embedding space of attribute learning,changing a specific attribute will causes other attributes to change,which limits the controllability and effectiveness of the retrieval results.In order to address this problem,this paper proposes an Attribute Preserving Disentangled Representation Learning(APDRL)network to realize interactive clothing retrieval.Firstly,the network learns a specific subspace for each attribute type of clothing image to ob-tain the attribute separation representation.Then,the joint loss function is used to optimize the network model.Finally,the manipu-lation vector modifies the specific attribute in the query image to the target attribute according to each attribute prototype embedded in the memory block.Testing on the DeepFashion dataset,the results show that APDRL method modify the target attribute while re-taining the fixed attribute of clothing,It has strong controllability and practicability,and is superior to other mainstream methods.关键词
交互式服装检索/解纠缠表示学习/DeepFashion/目标属性/损失函数Key words
interactive clothing retrieval/disentanglement representation learning/DeepFashion/target attribute/loss func-tion分类
信息技术与安全科学引用本文复制引用
黄晓菊,陈金广..基于属性解纠缠表示的交互式服装图像检索[J].计算机与数字工程,2025,53(3):829-834,6.基金项目
陕西省重点研发计划项目(编号:2020GY-122) （编号:2020GY-122）
陕西省教育厅科研计划项目(编号:21JP049)资助. （编号:21JP049）
