Virtual Dressing Room Application Using GANs

Abstract

In recent years, there has been a lot of interest in the idea of a virtual clothing try-on, which entails placing an image of a desired outfit on a mannequin. In order to give the user the idea that they are trying on the item, the standard try-on task attempts to provide a natural fit between the target clothing and the provided body. However, in real life, people may also care about how they "try on" in various stances. We offer a novel try- on setting that enables modification of both the garment and the model's posture in order to address this problem. To this purpose, we present a generative adversarial networks–based pose- guided virtual try-on technique (GANs). By initially predicting how the semantic layout of the reference image will change throughout the try- on process, it is possible to create a photorealistic try-on with a plethora of garment details. There are three individual parts to this. Initially, a reference image's semantic segmentation is used by a semantic layout generation module to progressively forecast the tried-on garment's desired semantic layout. A garments warping module then applies a second-order difference constraint to the created semantic layout in order to change clothing pictures while preserving the transformation's consistency over the course of training. Third, a module that can adapt to produce each semantic component of the human body from a variety of inputs (such as reference image, semantic layout, and distorted clothing). It outperforms cutting-edge generative models in the image-based virtual try-on task, according to our tests on the newly acquired dataset.