VisionApril 5, 2023Meta AI

Segment Anything

Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick

Abstract

We introduce the Segment Anything (SA) project: a new task, model, and dataset for image segmentation. Using our efficient model in a data collection loop, we built the largest segmentation dataset ever, with over 1 billion masks on 11 million licensed and privacy-respecting images. The model is designed and trained to be promptable, so it can transfer zero-shot to new image distributions and tasks.

Key Findings

  • 1Created the largest segmentation dataset with 1 billion masks on 11 million images
  • 2Introduced a promptable segmentation model (SAM) supporting points, boxes, and text prompts
  • 3Achieved strong zero-shot transfer to diverse segmentation tasks
  • 4Demonstrated a foundation model approach to computer vision segmentation
  • 5Released model and dataset as open source

Impact & Significance

SAM established the foundation model paradigm for image segmentation, making pixel-level understanding accessible through simple prompts. It transformed image editing tools, medical imaging, autonomous driving, and any application requiring object segmentation.

Read Full Paper