2D Cast Shadow Generation in E-commerce Image Using UNet Vision Transformer

in e-commerce, the visual presentation of products plays a crucial role in attracting customers and driving sales. One important aspect of product imagery is the generation of realistic cast shadows, which adds depth and enhances the overall visual appeal. In this paper, we propose the algorithm for 2D cast shadow generation in e-commerce using version of the Soft Shadow Network (SSN) modified by UNet Vision. The proposed model can handle complex lighting scenarios and produce realistic shadows with varying degrees of opacity. The UNet Vision Transformer helps to extract and encode the features of the object and its lighting conditions to generate the shadow maps. The model is employed to refine the generated shadows and ensure their consistency with the product images. The obtained results demonstrate that the use of the transformer has a positive effect on the quality of the trained network's prediction. The technology is integrated into the system developed by us which aim is to process the images and prepare e-commerce catalogue.