Generate 3D objects conditioned on text or images