Algorithm 1 Coordinate Attention (CA)

Input:  Input feature map $\mathbf{X}\in {\mathbb{R}}^{C\times H\times W}$  Output:  Output feature map $\mathbf{Y}\in {\mathbb{R}}^{C\times H\times W}$   1:# Global average pooling along height and width directions ${\mathbf{z}}_h={\displaystyle \frac{1}{W}}{\sum }_{j=1}^W\mathbf{X}(:,:,j)$    # Height descriptor ${\mathbf{z}}_h\in {\mathbb{R}}^{C\times H}$ ${\mathbf{z}}_w={\displaystyle \frac{1}{H}}{\sum }_{i=1}^H\mathbf{X}(i,:,:)$    # Width descriptor ${\mathbf{z}}_w\in {\mathbb{R}}^{C\times W}$   2:# Concatenate and transform via shared 1D convolution $\mathbf{z}=\mathrm{Concat}({\mathbf{z}}_h,{\mathbf{z}}_w)$    # $\mathbf{z}\in {\mathbb{R}}^{C\times (H+W)}$ $\mathbf{f}=\delta \left(\mathrm{Conv}1\mathrm{D}\right(\mathbf{z}\left)\right)$    # $\mathbf{f}\in {\mathbb{R}}^{{\displaystyle \frac{C}{r}}\times (H+W)}$   3:# Split and apply sigmoid activation ${\mathbf{f}}_h,{\mathbf{f}}_w=\mathrm{Split}(\mathbf{f},[H,W])$ ${\mathbf{g}}_h=\sigma \left({\mathrm{Conv}1\mathrm{D}}_h\left({\mathbf{f}}_h\right)\right)$    # Height attention ${\mathbf{g}}_h\in {\mathbb{R}}^{C\times H}$ ${\mathbf{g}}_w=\sigma \left({\mathrm{Conv}1\mathrm{D}}_w\left({\mathbf{f}}_w\right)\right)$    # Width attention ${\mathbf{g}}_w\in {\mathbb{R}}^{C\times W}$   4:# Apply attention weights to input features $\mathbf{Y}(i,j,:)=\mathbf{X}(i,j,:)\times {\mathbf{g}}_h(i,:)\times {\mathbf{g}}_w(j,:)$