Overview

SAM 2 extends Meta’s original Segment Anything Model from images to video. Given a click, box or mask on an object, it produces a precise outline and follows the object across video frames in real time, using a memory mechanism. Meta released the model and a large video segmentation dataset openly.

Highlights

  • Promptable segmentation of any object
  • Real-time tracking of objects through video
  • Works on objects not seen in training
  • Apache 2.0 license for the model
  • Released with the SA-V video segmentation dataset

Typical uses

Video editing and effects, data annotation, robotics perception, scientific imaging and medical research tools.

Things to consider

  • Segments objects but does not name or classify them by itself.
  • Very long videos and heavy occlusion remain challenging.
  • Production use needs GPU acceleration for real-time performance.

Specifications are taken from the provider’s official announcement, documentation or model card. Benchmark results depend on evaluation settings; see how AI benchmarks work.