Fele Benjamin ; Lampe Ajda ; Peer Peter ; Štruc Vitomir

Abstract

Image-based virtual try-on techniques have shown great promise for enhancingthe user-experience and improving customer satisfaction on fashion-orientede-commerce platforms. However, existing techniques are currently still limitedin the quality of the try-on results they are able to produce from input imagesof diverse characteristics. In this work, we propose a Context-Driven VirtualTry-On Network (C-VTON) that addresses these limitations and convincinglytransfers selected clothing items to the target subjects even under challengingpose configurations and in the presence of self-occlusions. At the core of theC-VTON pipeline are: (i) a geometric matching procedure that efficiently alignsthe target clothing with the pose of the person in the input images, and (ii) apowerful image generator that utilizes various types of contextual informationwhen synthesizing the final try-on result. C-VTON is evaluated in rigorousexperiments on the VITON and MPV datasets and in comparison to state-of-the-arttechniques from the literature. Experimental results show that the proposedapproach is able to produce photo-realistic and visually convincing results andsignificantly improves on the existing state-of-the-art.