Does multi-lr setting lead to some bugs?

last-one / Pytorch_Realtime_Multi-Person_Pose_Estimation

Pytorch version of Realtime Multi-Person Pose Estimation project

MIT License

216 stars 70 forks source link

Does multi-lr setting lead to some bugs? #40

Closed H-Liu1997 closed 4 years ago

H-Liu1997 commented 4 years ago

hi,last-one, I have tried to run your code,I found after remove multi-lr setting, the loss can be converged, if add multi-lr setting, the loss can't converge. and I add the bias decay = 0 in the multi-lr part, same as openpose, but I don't think this lead to the bug. have you faced this problem?