where is the gating operation?

[jcxu0]

Thanks for sharing the project,I have one problem, in the paper, gating operation follws the each [k,1,1] temporal convolutions, but in the code from S3DG_Pytorch.py， I can't find the gating operation...

class STConv3d(nn.Module):
    def __init__(self,in_planes,out_planes,kernel_size,stride,padding=0):
        super(STConv3d, self).__init__()
        self.conv = nn.Conv3d(in_planes, out_planes, kernel_size=(1,kernel_size,kernel_size),stride=(1,stride,stride),padding=(0,padding,padding))
        self.conv2 = nn.Conv3d(out_planes,out_planes,kernel_size=(kernel_size,1,1),stride=(stride,1,1),padding=(padding,0,0))

        self.bn=nn.BatchNorm3d(out_planes, eps=1e-3, momentum=0.001, affine=True)
        self.relu = nn.ReLU(inplace=True)
        
        self.bn2=nn.BatchNorm3d(out_planes, eps=1e-3, momentum=0.001, affine=True)
        self.relu2=nn.ReLU(inplace=True)
        
        nn.init.normal(self.conv2.weight,mean=0,std=0.01)
        nn.init.constant(self.conv2.bias,0)
    
    def forward(self,x):
        x=self.conv(x)
        #x=self.conv2(x)
        x=self.bn(x)
        x=self.relu(x)
        x=self.conv2(x)
        x=self.bn2(x)
        x=self.relu2(x)
        return x

[buaa-luzhi]

@jcxu0 i have the same question, do you know how to implement, thank you!

[syangdung]

i have the same question, i refer to other repositories and find that the common way is to utilize the squeeze and excitation to implement the feature gating mechanism.

[tobeatraceur]

i have the same question, i refer to other repositories and find that the common way is to utilize the squeeze and excitation to implement the feature gating mechanism.

Could you please share the links of the repos you mentioned? Thanks!