We propose a semantic-guided framework (FashionOn+) that generates image-based virtual try-on results with arbitrary poses. FashionOn+ contains three stages: (I) conducts the semantic segmentation to have the prior knowledge of body parts for rendering the corresponding texture in stage (II). (III) refines two salient regions, i.e., faceclothes, to generate high-quality results. With the novel architecture, we win first place in the Multi-pose Virtual Try-on Challenge in CVPR, 2020. Further, we tackle the low-resolution limitation (256x192)achieve high-resolution results (640x480).

"We move a step towards the real-world try-on scenario synthesizing try-on results with arbitrary poses. We propose a semantic-guided framework to deal with the issues caused by prior art techniques, e.g., pattern distortionlimited in one human pose. In terms of ISSSIM, our method surpasses the SOTA [1] by 18.919.8. In the Multi-pose Virtual Try-on Challenge in CVPR in 2020, our method won first placesurpassed the secondthird place in the A/B test by 8.624.7.

[1] Zheng et al.,“Virtually trying on new clothing with arbitrary poses,”in ACMMM, 2019."

Virtual try-on has great potentialities, especially with the COVID-19 pandemic. According to Statista, revenue in the Fashion segment is projected to reach US$878.3 billion in 2021. However, consumers tend to hesitate to decide whether to buy a specific garment since they are not sure whether the garment is suitable for them only based on one clothing image onlineeven an image with a model trying on the garment. To eliminate the uncertainty about whether the clothes are suitable for the consumer, our technique helps increase the conversion ratereduce the return rate.

1.欲媒合之產業領域：資訊與通訊、生活應用。2.欲媒合項目：技術合作、技術授權。