Publication

Abstract : Super resolution (SR) being one of the computer vision tasks with increasing applications in modern scenarios, several challenging factors are still prominent despite the numerous breakthroughs achieved in this field in recent years. Introduction of deep convolutional neural networks has brought a booming development to the existing SR techniques tackling many unsolved challenges. As an attempt to perform a relative analysis between currently used methods, this paper explores and establish the capability of Enhanced and Wide super resolution networks. These models are encompassed with improved residual networks with an aim to achieve a higher accuracy with reduced memory usage. The models trained with DIV2K dataset are evaluated using the T91 dataset and found to be showcasing a reliable performance in comparison with other cutting-edge methods devised for super resolution.