请教一下,loss用的什么呢,还有一些训练的trick可以讲一下吗
Closed this issue · 1 comments
interstellar-ai commented
RT
JPlin commented
We used FlowMatching loss and trained similarly to the original SD ControlNet, utilizing a dataset on the scale of tens of millions.
Closed this issue · 1 comments
RT
We used FlowMatching loss and trained similarly to the original SD ControlNet, utilizing a dataset on the scale of tens of millions.