Tracking the Invisible: Learning Where the Object Might Be

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 support