A Virtual Try-on Model with Enhanced Feature Representation Capability
Hui Ma, Zhuhua Hu, Yan Ping Zheng · 2022
When consumers choose to buy clothing online, virtual try-on technology can provide them with a better shopping experience. The optimization of virtual try-on technology not only helps consumers to evaluate the selected clothing, but also can improve the profit for merchants. However, the traditional virtual try-on technology has problems such as high cost, image distortion, and deviation of clothing style. In order to solve the above problems, this paper proposes a virtual try-on model with enhanced feature representation capability. Through the improved residual block of Squeeze-and-Excitation Networks (SENet) and the style encoding module introduced by the Pyramid Squeeze Attention (PSA) module, our model enriches the content and style information, strengthens the representation ability of features, and the reconstructed image preserves the more details. Compared with related work, we improve the structural similarity measure by 1.1% and the Inception Score by 10.1%. It is demonstrated that our model can reconstruct more accurate and realistic images.