No activation in BasicConv

[Callidior]

Hi, thanks for this repository! So far, SAM2UNet performs really well in my experiments.

I just came across some implementation detail that confused me a bit: The BasicConv block, of which sequences are used in the RBF module, has a ReLU child module but it is never used:

class BasicConv2d(nn.Module):
    def __init__(self, in_planes, out_planes, kernel_size, stride=1, padding=0, dilation=1):
        super(BasicConv2d, self).__init__()
        self.conv = nn.Conv2d(in_planes, out_planes,
                              kernel_size=kernel_size, stride=stride,
                              padding=padding, dilation=dilation, bias=False)
        self.bn = nn.BatchNorm2d(out_planes)
        self.relu = nn.ReLU(inplace=True)

    def forward(self, x):
        x = self.conv(x)
        x = self.bn(x)
        # SHOULD THERE BE `x = self.relu(x)` HERE ?
        return x

Was the activation omitted intentionally and, if so, what is the reason?

[xiongxyowo]

Hi, we exactly follow the RFB_modified design in PraNet and other popular salient/camouflaged object detection networks. For the original RFB in RFBNet, the ReLU is partially disabled. Unfortunately, the PraNet paper did not mention the motivation for this design difference about ReLU.

[Callidior]

Thanks for the quick answer! Glad to hear this is intentional :)