I am training an object detection model (retinanet) which has score threshold and nms threshold, as I understand other models do also. The documentation states these are part of postprocessing. Does postprocessing happen before the loss gets calculated, or is it something that happens when the model is actually performing inference?
What I am trying to understand is if these two values should be used as hyper parameters and tuned during training, OR are they something to play around with on your trained model.
Sorry if this is a really beginner question!