Text this: Leveraging learned monocular depth prediction for pose estimation and mapping on unmanned underwater vehicles