Паньков Кирилл Игоревич
Computer Modelling for Fashion E-Commerce
Today, a large number of people return clothes bought online. The main reasons are incorrectly chosen cloth’s size or a misconception about a product look. We consider solution to overcome both issues. Specifically, we developed a client-server application that enables a person to try clothes online on a virtual mannequin, duplicating client’s parameters, and also make recommendations on the optimal size of the selected cloth and recommend items, which might interest the potential consumer. To do so, a customer only has to provide to our system personal parameters and choose the apparel item from the store. For our system we use methods from both, machine learning and 3D modeling: human segmentation, garment warping, and convolutional neural networks for predictions. For training the neural network, we use DeepFashion dataset, which contains 800,000 diverse fashion images. As a result, we have a visual recommender system that can be integrated with online stores and work for benefit of fashion e-commerce companies and their consumers.