FRSFN: A semantic fusion network for practical fashion retrieval

Abstract

In recent years, research related to fashion has made remarkable progress, and the use of image content for fashion retrieval has become one of the effective approaches as well as research hot spots. However, it remains a challenging task due to the various contents contained in fashion images. This work presents a practical fashion retrieval method which puts emphasis on specific items. The Semantic Fusion Network of the method firstly extracts two kinds of features, which are the global features from the original query image and the item features. The item features are from the same query image semantically parsed before. Then the network fuses two kinds of features with the combination of color information. Finally, the similarity scores are calculated among features for retrieval. The experiments show that while remaining higher statistical retrieval results, our method grasps the detailed characteristics and items of the clothing and keeps a satisfying overall similarity in shape and color.