Affiliation:
1. Belarusian State University
2. Belarusian State University; United Institute of Informatics Problems of National Academy of Sciences of Belarus
Abstract
In the rapidly evolving landscape of e-commerce, the visual representation of products plays a pivotal role
in engaging consumers and driving conversion rates. This article introduces a new approach for image augmentation
that includes objects segmentation, dominant color determination, background replacement and realistic
shadow generation. These steps collectively contribute to the creation of augmented images that are used not only
in the electronic catalogues but enrich abilities of the neural networks with various and fortified training data. Developed
system allows to solve problems related to class imbalance and to enhance model generalization as well as to improve the quality of recognition.