Clothing attribute identification method based on migration significance prior information

Abstract

The invention relates to a clothing attribute identification method based on migration significance prior information. The clothing attribute identification method specifically comprises the followingsteps of S1, obtaining image data of garment images to perform attribute labeling and preprocessing; S2, inputting a saliency map into a saliency detection network to obtain the saliency map, and superposing the saliency map with the clothing image to form the garment image with saliency prior information; S3, executing the steps S1-S2 for each garment image, obtaining all garment images with saliency prior information, and inputting the garment images into a classification convolutional neural network for training until convergence; and S4, preprocessing a to-be-detected image in the step S1, then executing the step S2, obtaining the corresponding to-be-detected image with significance prior information, inputting the to-be-detected image with significance prior information into the trained classification convolutional neural network, identifying garment attributes, and outputting garment attributes in the to-be-detected image. Compared with the prior art, the method has the advantages that the generalization ability is high, the attribute recognition accuracy is improved, the investment of labor cost is reduced, and the like.

Description

technical field [0001] The invention relates to the fields of computer vision and deep learning, in particular to a clothing attribute recognition method based on transfer saliency prior information. Background technique [0002] Clothing attributes are the most direct basic information displayed to consumers by clothing products. It builds a matching relationship between consumers and clothing products, and directly guides consumers to purchase. In the past, in offline clothing stores, shopping guides often based on the purchase needs extracted by consumers, and then transformed them into corresponding clothing attributes based on experience, such as style, color, material and accessories, etc. recommend. Nowadays, online consumption has become a way of shopping for more and more people, and manual shopping guides have become impossible when hundreds of shopping needs are generated every second. Therefore, automatically recommending e-commerce products to potential or tar...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a method based on the prior information of migration significance in order to overcome the above-mentioned defects in the prior art that the prior information generated by manual labeling limits the application of the method and increases the difficulty of generalization of the method. Garment Attribute Recognition Method

Images