self.t = target_logit.mean() * 0.01 + (1 - 0.01) * self.t

[CloudWalking0]

Hello! In the paper, t should be calculated as:

It seems to be "self.t = target_logit.mean() * 0.99 + (1-0.99) * self.t" in the code.
But found "self.t = target_logit.mean() * 0.01 + (1 - 0.01) * self.t" in your code. I don't know why.
Thx.

[Geek-lixiang]

I have the same question!