Scene parsing

shape