Joint learning of visual attributes, object classes and visual saliency