Text-only Synthesis for Image Captioning