OccluMix: Towards De-Occlusion Virtual Try-on by Semantically-Guided Mixup

Image Virtual try-on aims at replacing the cloth on a personal image with garment (in-shop clothes), which has attracted increasing attention from multimedia and computer vision communities. Prior methods successfully preserve character of clothing images, however, occlusion remains pernicious effect for realistic virtual try-on. In this work, we first present comprehensive analysis occlusions categorize them into two aspects: i) Inherent-Occlusion: ghost former still exists in image; ii) Acquired-Occlusion: target warps to unreasonable body part. Based in-depth analysis, find that can be simulated by novel semantically-guided mixup module, generate semantic-specific occluded images work together facilitate training de-occlusion (DOC-VTON) framework. Specifically, DOC-VTON conducts sharpened semantic parsing person. Aided semantics guidance pose prior, various complexities texture are selectively blending human parts copy-and-paste manner. Then, Generative Module (GM) is utilized take charge synthesizing final learning jointly. comparison state-of-the-art methods, achieves better perceptual quality reducing effects.