LGM loss训练时出现loss为负值

YirongMao / softmax_variants

PyTorch code for softmax variants: center loss, cosface loss, large-margin gaussian mixture, COCOLoss, ring loss

253 stars 49 forks source link

Open liuge1008 opened 3 years ago

liuge1008 commented 3 years ago

LGM loss训练时大概20epoch 以后loss为负值，一直持续到结束，但是测试时效果还不错，请问loss的负值如何解决呢，期待您的回复！