SAM

Meta AI's foundational model for precise, zero-shot object segmentation in any image via point or text prompts.

Segment Anything Model (SAM) represents a shift in computer vision by treating image segmentation as a solvable prompt engineering task. Trained on the SA-1B dataset (1.1 billion masks across 11 million images), SAM executes real-time mask generation for unknown objects without additional training. It handles complex visual structures: fine-grained edges, overlapping layers, and varying scales. Developers integrate SAM via the Apache 2.0 license to automate labeling workflows, power augmented reality overlays, and drive autonomous spatial awareness.

https://segment-anything.com