real-time object pose estimation is necessary for many robot manipulation
algorithms. However, state-of-the-art methods for object pose estimation are
trained for a specific set of objects; these methods thus need to be retrained
to estimate the pose of each new object, often requiring