Text Promptable Surgical Instrument Segmentation with Vision-Language Models