training got NAN even if lr is set to 0

klintan / pytorch-lanenet

LaneNet implementation in PyTorch

MIT License

217 stars 60 forks source link

Open ginobilinie opened 4 years ago

ginobilinie commented 4 years ago

The instance segmentation loss is going to NAN even if the lr is set to 0.

klintan commented 4 years ago

Should be in a better shape now, but still some work left to do.