Interactive Multi-Channel Network for Fast Clothing Style Transfer

As an application of image style transfer, clothing style transfer has received considerable attention from academia and fashion industry due to its visual solutions for the generation and transfer of clothing styles. In this paper, we present a solution to interactive multi-style transfer for clothing that better preserves both structure and visual richness of clothing with fast generation. The core idea of our network model is to use multiple style channels to transfer the style of the labelled area and use the image fusion method to generate personalized clothing styles. Users can control the spatial distribution of clothing stylized textures and colors through the semantic annotation. We demonstrate the effectiveness and superiority of our method through extensive comparisons with state-of-the-art algorithms.