Segment Anything Model 2 (SAM 2) by Meta AI

SAM 2 is Meta AI’s next-generation, unified segmentation model engineered for both images and videos. Featuring a cutting-edge transformer architecture with a streaming memory module, SAM 2 delivers real-time, promptable segmentation with exceptional accuracy and speed.
From dynamic video tracking to precise object segmentation in still images, SAM 2 empowers developers and businesses across industries—from video editing and augmented reality to medical imaging and autonomous vehicles.

Key Features of SAM 2 include:

• Unified Segmentation: Seamlessly processes both images and videos with one model.
• Real-Time Performance: Achieves rapid segmentation at up to 44 frames per second for video and is 6x faster on images.
• Advanced Prompting: Supports interactive segmentation using points, bounding boxes, or masks, with robust zero-shot generalization.
• Memory-Enhanced Tracking: Utilizes a streaming memory module to maintain consistent object tracking across video frames.

To learn more about SAM 2, visit the official SAM 2 page.