Deep Learning for Fashion Analysis

Abstract

Using semantic segmentation algorithms to automatically classify each pixel of an image, could imply great benefits for the fashion industry but also for the use of society. Semantic segmentation algorithms can for example be used within the fashion industry by predicting trends on social media or by robots within health care to help people get dressed. However, performance of semantic segmentation algorithms are dependent on a large amount of annotated training data. The fashion industry in particular, is an area where machine learning algorithms are not as well developed as in many other fields, and the amount of training data is thus limited. Therefore, the purpose of this thesis was to investigate if it is possible to use deep learning for generative modeling to increase the amount of training data in the fashion domain. The results showed that it is possible to use generative adversarial networks (GANs) to generate pairs of images and corresponding pixel wise annotations.