The Segment Anything (SA) project introduces a new task, model, and dataset for image segmentation. The SA model, called Segment Anything Model (SAM), is designed to be promptable, enabling zero-shot transfer to new image distributions and tasks via prompt engineering. The project builds the largest segmentation dataset to date, SA-1B, containing over 1 billion masks on 11 million images. SAM is trained on a diverse, large-scale dataset using a data engine that iteratively collects and improves data. The model is capable of generating segmentation masks from various prompts, including points, boxes, and text. SAM is evaluated on numerous tasks and shows strong performance, often competitive with or superior to prior fully supervised results. The SA-1B dataset is released to foster research into foundation models for computer vision. The project also includes a responsible AI analysis, showing that SAM performs similarly across different groups of people. SAM is released under a permissive open license at https://segment-anything.com. The model is capable of zero-shot transfer to various tasks, including edge detection, object proposal generation, instance segmentation, and text-to-mask prediction. The SA project demonstrates the potential of foundation models in computer vision, enabling a wide range of downstream tasks through prompt engineering.The Segment Anything (SA) project introduces a new task, model, and dataset for image segmentation. The SA model, called Segment Anything Model (SAM), is designed to be promptable, enabling zero-shot transfer to new image distributions and tasks via prompt engineering. The project builds the largest segmentation dataset to date, SA-1B, containing over 1 billion masks on 11 million images. SAM is trained on a diverse, large-scale dataset using a data engine that iteratively collects and improves data. The model is capable of generating segmentation masks from various prompts, including points, boxes, and text. SAM is evaluated on numerous tasks and shows strong performance, often competitive with or superior to prior fully supervised results. The SA-1B dataset is released to foster research into foundation models for computer vision. The project also includes a responsible AI analysis, showing that SAM performs similarly across different groups of people. SAM is released under a permissive open license at https://segment-anything.com. The model is capable of zero-shot transfer to various tasks, including edge detection, object proposal generation, instance segmentation, and text-to-mask prediction. The SA project demonstrates the potential of foundation models in computer vision, enabling a wide range of downstream tasks through prompt engineering.