Dress-On online model base on Deep Learning and Computer Visio

Abstract

Image-based virtual try-on seeks to fit target clothing into an image of a person. Especially, garment warping is a crucial step for aligning the desired garment with the body portions in the image. The previous model [(2)] typically applies a local appearance flow method for warping modules. The results of them are mostly harmful to difficult body poses and large misalignments between person and garment images. To minimize this limitation, we proposed a novel global appearance flow estimation model. This allows us to overcome the aforementioned difficulties by utilizing a global style vector to convey a whole-image context. Also, a flow refinement module is presented to add local context so that the StyleGAN flow generator can pay more attention to local garment deformation