Identity-Guided Face Generation with Multi-modal Contour Conditions