Dense Object Nets: Learning Dense Visual Object Descriptors By and For Robotic Manipulation

We present Dense Object Nets, which build on recent developments in self-supervised dense descriptor learning, as a consistent object representation for visual understanding and manipulation.