Panoptic-Level Image-to-Image Translation for Object Recognition and Visual Odometry Enhancement