Integrating 3D And 2D Representations For View Invariant Object Recognition