Open vadimkantorov opened 4 years ago
Could you please comment on the particular way of decaying the BatchNorm momentum parameter for every mini-batch during BatchNorm parameters update https://github.com/pytorch/contrib/blob/master/torchcontrib/optim/swa.py#L305 ?
(As far as I understand, BatchNorm momentum is usually constant)
Thanks!
Could you please comment on the particular way of decaying the BatchNorm momentum parameter for every mini-batch during BatchNorm parameters update
https://github.com/pytorch/contrib/blob/master/torchcontrib/optim/swa.py#L305 ?
(As far as I understand, BatchNorm momentum is usually constant)
Thanks!