Deep Learning based Swapping Generative Framework for Rapid Cloth Retrieval

Abstract

Abstract Modern life is fast-paced, and every people is very busy with their daily routines. The online shopping option in E-commerce is a great time-saver in such a scenario. Moreover, it is crucial to extract accurate product features in order to search, recommend, classify, and retrieve images based on fashion queries.To overcome the forementioned challenges, a novel cloth swapping GAN based fashion retrieval has been introduced for rapid retrieval of relevant fashion based on the user query. Initially, to reduce the computational time, GrabCut is used to remove the background of the cloth images.The Cloth encoding decoding-based parsing Network is introduced to segment the bottom and top of the cloth. Then, the separated cloth region is fed into the GAN based on the user preference. The threshold neural network (TNN) is integrated with gates for efficient feature extraction in a small fraction of time. The feature extraction process is performed based on the feedback of the user. The extracted features such as dress length (long, medium, short), dress sleeve (sleeveless, full sleeve, half sleeve), and dress pattern (designs, dots, straights) are used to retrieve the relevant clothes for the users based on the query from the online shops. The proposed model achieves atotal accuracy of 99.29%. The proposed cloth retrieval system enhances the total accuracy by 14.24%, 8.75%, and 23.55% better than Alexnet, cGAN, and CNN, respectively.