自定义数据集上训练过程中出现loss=='nan'

[serendipity999]

默认参数未变更，在自己的数据上微调，调试看到某一时刻output为nan，然后loss计算出错，请问会是什么原因呢？

[SAWRJJ]

我是在训练的过程中出现了loss = ‘inf’的情况 朋友你的情况最后是怎么解决的

[bubbliiiing]

试试全解冻 ，不冻结

[abhigoku10]

@bubbliiiing even i am facing the same issue , can u please elaborate the solution

[bubbliiiing]

I want to, but I don't know how to reproduce it

[abhigoku10]

@bubbliiiing i have attached the log file in this comment pls check even if i am having loss values for all the different loss yet total loss is nan/inf yolact_pytorch_bubbling_colab_log.txt

[bubbliiiing]

Maybe the list of mean is empty

[ruonan126]

请问loss = 'inf'这个问题有解决方法吗？

[Andrew-Ma-2001]

Same problem