Loss is NaN - Githubissues

kuangliu / pytorch-retinanet

RetinaNet in PyTorch

992 stars 250 forks source link

Closed Mendel1 closed 7 years ago

Mendel1 commented 7 years ago

When I'm training the retinanet,cls_loss is easy to become NaN.Does anyone know the reason?

loc_loss: 0.084 | cls_loss: nan | train_loss: nan | avg_loss: nan

kuangliu commented 7 years ago

See #6. I'll update the loss soon.