Research Of Virtual Try-on Technology Based On Two-dimensional Image

The emergence of virtual try-on technology enables people to try on clothing without leaving home,which is convenient,fast,and efficient,and has great practical significance and research value.Virtual try-on based on two-dimensional image refers to trying on the target clothing on a human image to generate the final fitting image.Users do not need to personally try on the clothing to obtain the effect image after the try on,solving the difficulty that users cannot try on when shopping online,and bringing new vitality to the clothing industry.Currently,there are some problems about virtual try-on technology,such as blurred image of the human,loss of body parts,and clothing that cannot be correctly deformed according to the posture of the human images.To address the above issues,this paper studies virtual try-on technology based on two-dimensional images to generate higher quality fitting images.The main research content and innovation points of this article are as follows:(1)This paper proposes a virtual try-on method A-VITON based on the improvement of Flow-Style-VTON network structure.In order to generate more realistic fitting images under the guidance of human parsing,the clothing deformation and fitting stages based on the parser-free model are removed,and the four-stage model is changed to a two-stage model.In order to make the model converge better,this paper embeds Res Net V2 residual blocks into U-Net,and the residual blocks are pre-activated.In order to generate high-quality images,this paper adds a CBAM module as an attention mechanism to learn important information during network down-sampling.(2)Currently,the virtual try-on model only focuses on replacing the original clothing with the clothing to be tried on,without considering the user’s layered fitting needs.In order to further improve the virtual try-on experience,this paper proposes a layered virtual try-on method to provide consumers with more diverse fitting services.On the virtual try-on method A-VITON proposed in this paper,the clothing deformation module is retained,and the clothing fitting module is changed to retain some of the original clothing to achieve layered try-on.(3)Complex backgrounds can interfere with the feature extraction capabilities of the network,resulting in poor performance of the model in processing human images with complex backgrounds.Therefore,in order to reduce the interference of complex backgrounds on fitting results,this paper proposes for the first time a virtual try-on method with backgrounds,which can generate high-quality fitting images while retaining the original image background.On the basis of the A-VITON model,a background removal module is added in the preprocessing stage.This module uses the U~2-Net network model to achieve target edge detection,highlighting the target area,and in the generation of the try-on stage,a background fusion module is added to integrate the background into the try-on image to generate the final try-on result.The virtual try-on method A-VITON proposed in this paper is based on the VITON dataset for training and testing.The qualitative comparison results with other mainstream models show that the model in this paper is more capable of generating high-quality fitting results,and can well preserve the texture and pattern of clothing.And when the human body’s limbs block clothing,it can also generate good fitting effects.The quantitative comparison results show that the model in this paper has achieved improvements in both SSIM and FID evaluation indicators,and is more robust,whether compared with parser-based models or parser-free models.