A Joint Parsing System For Visual Scene Understanding