SLAM and semantic reconstructions

(i)evaluate the methods using different metrics, such as efficiency in speed (inference or training) or data, and (ii) consider the real-world applications beyond just semantic labelling. In reality this system might be used by a robot to navigate or manipulate its environment, and one method might be better suited to this than the other. For example, I personally think that a height-map approach might struggle for manipulation tasks, regardless of how good it is at recognising the particular object you want the robot to manipulate.

Research area: Computer Vision