Why update_lr() ?
oneHuster opened this issue · 1 comments
oneHuster commented
I was so confused about using self.update_lr(cur_nimg)
in train.py train.py#L281 which may produce negative learning rate coefficient during the training process.
I have no idea about why we should use this func --- a func just make learning rate fluctuating.
github-pengge commented
This is a trick taken from official code.