Objects are usually embedded into context. Visual context has been successfully used in object detection tasks, however, it is often ignored in object tracking. We propose a method to learn supporters which are, be it only ...