3D virtual fitting network (3D VFN)

Leong, Wei Wen (2022) 3D virtual fitting network (3D VFN). Final Year Project, UTAR.

Preview

PDF Download (10Mb) | Preview

Abstract

Ever since the pandemic has kicked in, people have started adopting shopping from home. This has led to tremendous impact on the clothing and garment industry that causes operation shutdowns and store closures. Recently, to mitigate the issue addressed, research on Virtual Try-On (VTO) technologies to be implemented in the virtual fitting rooms (VFRs) have drawn significant attention. The existing VFRs in the market implement VTO solutions relying on deep generative models with an end to-end pipeline, from feature extraction to garment warping and refinement. The VTO solutions can be categorised into two, which are 2D and 3D networks. 3D VTO solutions have an enormous commercial potential in the retail fashion market as the technology has been proven effective for providing a photo-realistic and detailed try on result. However, the existing 3D VTO solutions principally rely on annotated human body shapes or avatars, which are impractical and unrealistic. By integrating the technologies embedded in both the existing 2D and 3D VTO solutions, this project proposes a VTO solution which relies on the geometric settings in the 3D space, namely the 3D Virtual Fitting Network (3D VFN), which produces a warped garment output image with texture and details preserved. The 3D VFN solely relies on 2D RGB images, which takes in a garment image and a single-person human image as inputs.

Actions (login required)