Meaningful validation metric for rotated object detection

What is a good validation metric to use for object detection with rotated bounding boxes?

I am working with satellite images so I care about correctly predicting the “head” of the box.

I am regressing for the rotation angle during training but I am unsure how the F1 score takes this into account since it measure the overlap of my ground truths and predictions without regarding their orientations.

Thank you in advance.