This project focuses on Image Enhancement through Super Resolution, leveraging a model introduced in the paper "Fast, Accurate, and Lightweight Super-Resolution with Cascading Residual Network." The primary challenge addressed in this research is the high computational cost associated with existing methods for Single Image Super-Resolution tasks. To mitigate this issue, the paper proposes an efficient and lightweight model that offers performance comparable to state-of-the-art (SoTA) techniques. The key innovation lies in the adoption of a cascading mechanism within a residual network. These mechanisms facilitate the efficient transfer of information to higher layers, expediting the learning process. Furthermore, the paper introduces residual-E blocks, leveraging group convolution, to reduce the number of computational operations (MultiAdds). In comprehensive evaluations on standard datasets such as Set5 and Urban100, the proposed model consistently demonstrates performance levels akin to SoTA methods. This project exploits the insights from the paper to implement an advanced solution for enhancing image quality through super-resolution.
Paper - https://arxiv.org/abs/1803.08664 [Fast, Accurate, and Lightweight Super-Resolution with Cascading Residual Network]