Toward Semi-Supervised Graphical Object Detection in Document Images