Видалення тіней із зображення за допомогою генеративних змагальних мереж та навчання без учителя

Abstract

This material presents the solution for shadow removal task using generative adversarial networks. Our approach is trained in unsupervised fashion which means it does not depend on time-consuming data collection and annotation. This together with training in a single end-to-end framework significantly raises its practical relevance. Taking the existing method for unsupervised image transfer between different domains we researched its applicability to the shadow removal problem. By exploiting attention modules and multi context feature aggregation using dilated convolutions our method gives significant results compared to existing solutions in the field.