mAP for object detection on test split

[AashishKumbashi]

I understand that after every epoch, rastervision helps in calculating the mAP on the validation dataset. However, later when analysing the predictions, I have only precision, recall and F1 score.

Is there a way to somehow obtain the mAP scores for the test dataset?

[AdeelH]

You are right. That's some missing functionality. To fix that, we would need to adapt the mAP calculation code from here to here.

A workaround might be to instantiate a Learner with the test dataset and the trained weights and then call Learner.eval_model('test').