The parameters of loss are always similarity during training

Mikoto10032 / AutomaticWeightedLoss

Multi-task learning using uncertainty to weigh losses for scene geometry and semantics, Auxiliary Tasks in Multi-task Learning

Apache License 2.0

575 stars 81 forks source link

The parameters of loss are always similarity during training #11

Open Leputa opened 2 years ago

Leputa commented 2 years ago

I have 3 losses, such as """ tensor([0.9926, 0.9926, 0.9927], requires_grad=True) tensor([0.9908, 0.9908, 0.9909], requires_grad=True) tensor([0.9873, 0.9873, 0.9873], requires_grad=True) """

long123524 commented 2 years ago

我也是这样的，请问你解决了吗？

timerobin commented 1 year ago

I meet the same problem two

zhiqihuang commented 5 months ago

Same issue. Any idea?