基于深度卷积网络的同款商品图像检索研究

Research on Large Scale Same Style Commodity Image Retrieval Based on Deep Convolution Neural Network

使用深度卷积神经网络来进行大规模的同款商品图像检索研究,同时设计一种可利用多种类标信息来进行神经网络训练的网络结构,并与传统的Bo F图像检索框架进行对比。相较于传统的方法,基于深度卷积神经网络的检索精度有较大幅度的提高。

Designs and trains a new structure of deep convolution neural network using multi-labeled image. Uses it to do same style commodity image retrieval. Compared with traditional bag of features method, it gets a much higher MAP score.