About the loss

[ZHUANGMINGXI]

When I train, inst_loss is always nan, and only seg_loss has a value by chance. Is the above problem normal?

[xuxw98]

Yes, it's okay. There are some errors in the log but we just ignore that:)