Virtual Cloth Warping using Deep Learning

Abstract

Virtual Try-On is a technology that can realistically clothe an individual virtually, it transfers a clothing image onto a target person's image. This is attracting attention from the industrial and research centers and can make the in-store experience achievable. The 2D image-based and 3D model-based methods developed recently have their own benefits and limitations. This paper describes the development of a 2D image-based Virtual Try-On Clothing system. Our solution comprises major modules: Human representation which is pose estimation using OpenCV and human parsing using Self Supervised Joint Body Parsing and Pose Estimation Network (SS-JPPNet) and the Try-On module which utilizes Cloth Warping Module (CWM) and Cloth Fusion Module (CFM) to generate the final try-on output. The technologies that are used for CWM is Convolutional Neural Network and for CFM is Generative Adversarial Network. Our application as of now supports only upper body clothing (tops, t-shirts, etc.). A graphic user interface is created where one can virtually try on clothes by uploading their picture and selecting the clothing item to try on, bringing the shopping experience to one's doorstep.