I think there is a wrong in Focalloss.py

[taylover-pei]

You write the focalloss just like this:

By doing this, the logpt is a scale not a tensor. If the batchsize is greater than 1, I think it is wrong here.
An I right?