the train loss is nan

[MrYANG23]

hi! thanks for your work ,when i train this repo i get nan loss no matter use the resnet18 or using the resnet50. can you give me some advices,thank you !

[PuAnysh]

Me Too.this problem will occur while i use loss_coor in final loss.

[wangziren1]

Maybe, dropping learning rate by ten times or five times helps.