Co-segmentation for visual object tracking