Table 5 Pseudocode of the multi-scale channel attention module (MS-CAM).

Input: Feature map \({\text{X}} = [{\text{x}}_{1} ,{\text{x}}_{2} , \ldots ,{\text{x}}_{C} ] \in {\mathbb{R}}^{C \times H \times W}\)

Parameters:

           \(\mathcal{B}\) denotes the batch normalization (BN) layer

           \(\delta\) denotes the linear rectifier function (ReLU)

           \({\mathcal{V}}\) denotes the view function

           \(\sigma\) denotes the sigmoid function

           \(\oplus\) denotes the additive function

           \(\otimes\) denotes the multiplicative function

1

// Local Attention

2

\(Y_{1} = \mathcal{B}({\text{PWConv}}_{1} (X))\)

3

\(Y_{2} = \delta (Y_{1} )\)

4

\({\text{L}} (X) = \mathcal{B}({\text{PWConv}}_{2} (Y_{2} ))\)

5

// Efficient Channel Attention

6

\(Y_{1} = {\mathcal{V}}({\text{Avgpool}} (X))\)

7

\(Y_{2} = {\mathcal{V}}(\sigma (1DConv(Y_{1} )))\)

8

\({\text{E}} (X) = Y_{2} \otimes X\)

9

// MS-CAM

10

\({\text{M}} ({\text{X}} ) = X \otimes \sigma ({\text{L}} (X) \oplus {\text{E}} (X))\)

Output: the outputs after the filters as \({\text{M}} ({\text{X}} ) \in {\mathbb{R}}^{C \times H \times W}\)