Image Captioning With Multi Context Synthetic Data