no relu before the global pool?

[hengck23]

in the classification model of nextvit.py:

why is there no relu before the global pool?
the batchnorm will produce both +ve and -ve values.

    def forward(self, x):
        x = self.stem(x)
        for idx, layer in enumerate(self.features):
            if self.use_checkpoint:
                x = checkpoint.checkpoint(layer, x)
            else:
                x = layer(x)
        x = self.norm(x)
        x = self.avgpool(x)
        x = torch.flatten(x, 1)
        x = self.proj_head(x)
        return x