LMCap: Few-shot Multilingual Image Captioning by Retrieval Augmented Language Model Prompting