Representational frames in dynamic scene annotation