region awareness loss

[diksham-flwls]

https://github.com/AaronComo/LipFD/blob/c2a73242c36f604ae7377d77b94c00faf9d942fd/models/LipFD.py#L40

The loss value needs to be averaged across the time dimension as well as otherwise this will blow up the loss value for RA Loss and overpower the BCE loss. Unless I am missing sth here...