learning hierarchical features for scene labeling published presentations and documents on DocSlides.