The difference between the code and the legend in the paper

[wugggfff]

I noticed that in the code.In the MSCASpatialAttention,There is the following code def forward(self, x): """Forward function."""

    shorcut = x.clone()
    x = self.proj_1(x)
    x = self.activation(x)
    x = self.spatial_gating_unit(x)
    x = self.proj_2(x)
    x = x + shorcut
    return x

as we can see，when the X go through the 1x1 convolution、activation（GELU）、MSCAAttention and another 1x1 convolution，X add its shortcut！

But in the legend of the Attention( corresponding to the MSCASpatialAttention part in the code)，it doesn‘t draw this operation similar to residual linking

So I would like to ask, is the operation here not shown in the diagram

[wugggfff]

in addition,the expansion ration of FFN = [8,8,4,4] in the paper, but in the code,the mlp_ratio = [4,4,4,4] in the class MSCAN can you explain it please?