RAINet: Rain‐Aware Image Deraining Network with Multi‐Stage Image Decomposition

Rain degrades outdoor image quality, making deraining a challenging task due to varying rain density and direction. We propose a novel Rain‐Aware Image Deraining Network (RAINet), leveraging a Transformer‐based architecture to strategically separate attention and MLP modules to decompose rain layers and clean backgrounds. It leverages global and local rain‐aware attention to capture rain features, while dilated convolutions aggregate contextual background before MLP processing. This multi‐stage strategy efficiently produces clear, rain‐free images, outperforming state‐of‐the‐art models across diverse conditions.