ClothingOut: a category-supervised GAN model for clothing segmentation and retrieval

This paper presents a new framework, ClothingOut, which utilizes generative adversarial network (GAN) to generate tiled clothing images automatically. Specifically, we design a novel category-supervised GAN model by learning transformation rules between clothes on wearers and clothes that are tiled. Our method features in adding category attribute to a traditional GAN model. For model training, we built a large-scale dataset containing over 20,000 pairs of wearer images and their corresponding tiled clothing images. The learned model can be straightforwardly applied to video advertising and cross-scenario clothing image retrieval. We evaluated our generated images which can be regarded as the segmentation from the wearer images from two aspects: authenticity and retrieval performance. Experimental results demonstrate the effectiveness of our method.