{Yu-Kun Lai Heejune Ahn Paul Rosin Thai Thanh Tuan Matiur Rahman Minar}

Abstract

Recently proposed Image-based virtual try-on (VTON) approaches have several challenges regarding diverse human poses and cloth styles. First, clothing warping networks often generate highly distorted and misaligned warped clothes, due to the erroneous clothing-agnostic human representations, mismatches in input images for clothing-human matching, and improper regularization transform parameters. Second, blending networks can fail to retain the remaining clothes due to the wrong human representation and improper training loss for composition mask generation. We propose CP-VTON+ (Clothing shape and texture Preserving VTON) to overcome these issues, which significantly outperforms the state-of-the-art methods, both quantitatively and qualitatively.