Published Paper Details:

In online shopping, People would like to know how they look in a particular dress they tend to buy. To bring closer between in-store (offline) and online shopping and provide a realistic shopping experience to the consumers, a system is required to predict how people look in a particular garment that they tend to buy. Traditional approaches to make such a system were dealing with computer graphics where a 3D avatar of the person is created and various clothes are visualized on him/her. The usage of such approaches limits practical applications due to high computation and hardware requirements. This Project aims to build a fitting system such that, it visualizes people wearing new clothes without any 3D information, only using Images, in the same pose as they are standing in. This Project is composed of two modules one for changing the shape of cloth given in the input according to the personality and another module to attach this cloth onto the individual body. Using Efficient CNN based networks, which learns to predict the parameters for changing shape via spatial transformation is used in the first module. The shape changed cloth is passed on to the next module which is based on U-net architecture that predicts the mask where to attach the cloth, using this information the cloth is attached onto the body for generating final output images. The system is able to generate natural and rich images without much loss.