基于多尺度融合增强的服装图像解析方法

Clothing Image Parsing Method Based on Multi-scale Fusion Enhancement

摘要:

基于卷积神经网络中的各个层次特征，提出了一种基于多尺度融合增强的服装图像解析方法。通过融合增强模块，在考虑全局信息的基础上对包含的语义信息和不同尺度特征进行有效融合。结果表明：在Fashion Clothing测试集上的平均F1分数达到60.57%，在LIP（Look Into Person）验证集上的平均交并比（mean intersection over union，MIoU）达到54.93%。该方法可以有效地提升服装图像解析精度。

Abstract:

By using the features of each level in convolutional neural network， a clothing image parsing method based on multi-scale fusion enhancement was proposed. Through the fusion enhancement module， the semantic information and the features in different scales were effectively fused with the consideration of global information. The results show that the average F1 score on the Fashion Clothing test set reaches 60.57%， and the mean intersection over union（MIoU） on the Look Into Person（LIP） validation set reaches 54.93%. The method can effectively improve the accuracy of clothing image parsing.