基于通道注意力的多模态服装兼容性学习

Multimodal Clothing Compatibility Learning Based on Channel Attention

针对服装图像特征提取不全面和服装兼容性难以建模等问题,提出了一种基于通道注意力的多模态服装兼容性模型ECA-RMCN。在特征提取网络CNN的残差模块上引入高效通道注意力模块ECA-Net来增强服装低级和高级等重要特征,抑制无效特征。采用组合损失函数处理服装正负样本不均衡的问题,达到更好的搭配效果。在公共的Polyvore数据集进行对比实验来验证模型有效性。实验结果表明,该算法对服装的兼容性预测和搭配性能优于其他方法,有很好的应用价值。

Aiming at the problems of incomplete feature extraction of clothing images and difficult modeling of clothing compatibility and so on,a multimodal clothing compatibility model ECA-RMCN based on channel attention is proposed.The high-efficiency channel attention module ECA-Net is introduced on the residual module of the feature extraction network CNN to enhance important features such as low-level and high-level clothing,and suppress invalid features.The combined loss function is used to deal with the problems of unbalanced positive and negative samples of clothing to achieve better matching effect.Comparative experiments are performed on the public Polyvore dataset to verify the effectiveness of the model.The experimental results show that the algorithm is better than other methods in the compatibility prediction and matching performance of clothing,and it has good application value.