Can GeneralizedRCNN forward pass return both loss AND inference results?

[take5v]

Very often it is desirable to compute both loss function and different metrics using inference results on a validation set. For this, in a current version of GeneralizedRCNN, two methods should be called separately (forward and inference) requiring two network passes. Adding inference results to return values will increase validation time when both loss and metrics are computed.

[CloudyDory]

I would also like to see this functionality. However you can always look into the code and write a custom function that calculates both losses and inference results.