chunked/nltk_chunking/model_doc_sam/chunk_8.txt

The model predicts much better results if input 2D points and/or input bounding boxes are provided
You can prompt multiple points for the same image, and predict a single mask.