Overview & Capability
Meta's foundation model for promptable visual object segmentation in images and real-time video streams with interactive point, box, and mask tracking.
Key Highlights
- Real-time video object segmentation at 44+ FPS
- Memory attention architecture for tracking occluded objects across video frames
- Zero-shot generalization to unseen visual domains
Robbie's Independent Take
Verified: 2026-10-11
Evaluated independently for utility and transparency. Test against a controlled baseline workload before integrating into mission-critical production pipelines.
Alternatives & Replaces
- Manual rotoscoping & computer vision annotations