anyone knows why _freeze_stages() starts from block[0]?

[abc5z7]

thanks for your great code. But i have a question about freeze parameters in mmpose/models/backbones/vit.py line 273

    def _freeze_stages(self):
        """Freeze parameters."""
...
        for i in range(1, self.frozen_stages + 1):
            m = self.blocks[i]
            m.eval()
            for param in m.parameters():
                param.requires_grad = False
...

when i need freeze all the parameters but the code shows it starts from index 1. As block[0] is alway require gradient.

Is block[0] have any other reasons? This setting is somehow training strategy or something else?

thanks again!