WarpClothingOut: A Stepwise Framework for Clothes Translation From the Human Body to Tiled Images

Abstract

With the increasing popularity of online shopping, searching for products with images for item retrieval has gradually become an effective approach. This trend is especially evident in the fashion industry. In common media, clothing items are usually worn on the human body. They can be straightforwardly segmented from the source media by utilizing detection or parsing algorithms. However, this may be deleterious to retrieval performance due to distortion, occlusion, and different backgrounds. In this article, a stepwise translation framework using generative adversarial network and thin plate spline is developed to transfer human body images to tiled clothing images, which can be directly used for clothing retrieval. Experimental results demonstrate the effectiveness of the resultant tiled images produced from our framework in comparison to other extant methods.