Training retinanet_resnet50_fpn model

Hi everyone!
Im trying to train this model: torchvision.models.detection.retinanet — Torchvision 0.11.0 documentation.

During inference (model.eval()) the model returns some “scores”.

  1. How can i obtain the loss values from the model in evaluation mode? Can calculate the loss = (1 - score)?
  2. Do this scores relate to classification or for bbox_regression?

Thank you!