A High-resolution Image-based Virtual Try-on System in Taobao E-commerce Scenario

Zhilong Zhou, Shiyao Wang, Tiezheng Ge, Yuning Jiang

Abstract: On an e-commerce platform, virtual try-on not only improves consumers' shopping experience but also attracts more consumers. However, the existing public virtual try-on dataset can't be generalized to Taobao fashion items because of low resolution, racial differences, and small dataset size. In this work, we build a large-scale and high-resolution virtual try-on dataset for Taobao e-commerce. Based on the dataset, we propose a virtual try-on system to generate high-resolution and attractive virtual try-on images without time-consuming image preprocessing. The system has covered more than 10000 fashion items and produced millions of virtual try-on results in Taobao e-commerce scenario.