Algorithm 1 CMAWRNet Image Restoration Algorithm

Require: Input image $I\in {\left[0,1\right]}^{W\times H\times 3}$ Ensure: Restored image $I_{\mathrm{clean}}$ Initialization: Set parameters ${\gamma }_t,{\gamma }_s,\gamma$, and patch size $\Omega$ Compute guidance map: $\mathbf{G}\leftarrow {\left|\nabla I\right|}^{{\gamma }_t}$ $⊳$ Equation (5) Compute structure map: $S_0\leftarrow {\displaystyle \frac{1}{\left|\Omega \right|}}\sum _{\left(i,j\right)\in \Omega }{\left|\nabla I\right|}_{i,j}^{{\gamma }_s}$ $⊳$ Equation (6) Compute texture component: $T_0\leftarrow I\oslash S_0$ $⊳$ Since $I=S_0\circ T_0$ Refine components using DNet: $S\leftarrow \mathrm{DNet}\left(S_0\right)$ $⊳$ Refined structure $T\leftarrow \mathrm{DNet}\left(T_0\right)$ $⊳$ Refined texture Clean components: $S_{\mathrm{clean}}\leftarrow \mathrm{TNet}\text{-}\mathrm{H}\left(S\right)$ $⊳$ Clean structure component $T_{\mathrm{clean}}\leftarrow \mathrm{TNet}\text{-}\mathrm{S}\left(T\right)$ $⊳$ Clean texture component Estimate latent variable from structure: $M\leftarrow \mathrm{Conv}1\mathrm{x}1\left(S_{\mathrm{clean}}\right)$ ${\tilde{S}}_{\mathrm{clean}}\leftarrow M\circ S_{\mathrm{clean}}$ ${\widehat{S}}_{\mathrm{clean}}\leftarrow S_{\mathrm{clean}}^{\gamma }$ $⊳$ Gamma correction Compute attention maps using FNet: $M_S\leftarrow \sigma \left(W_2^{\left(S\right)}\mathrm{ReLU}\left(W_1^{\left(S\right)}\left[S_{\mathrm{clean}},I\right]\right)\right)$ $M_T\leftarrow \sigma \left(W_2^{\left(T\right)}\mathrm{ReLU}\left(W_1^{\left(T\right)}\left[T_{\mathrm{clean}},I\right]\right)\right)$ Fuse components: $S_{\mathrm{fused}}\leftarrow M_S\otimes S_{\mathrm{clean}}$ $T_{\mathrm{fused}}\leftarrow M_T\otimes T_{\mathrm{clean}}$ $I_{\mathrm{clean}}\leftarrow S_{\mathrm{fused}}\circ T_{\mathrm{fused}}$ return $I_{\mathrm{clean}}$