Detecting Object Boundaries Using Low-, Mid-, and High-level Information