Key Info
A robotics demonstration combines SAM 3.1 with Meta's DINOv3 to track surgical tools, distinguishing between near-identical twins even when one leaves the scene for 15 seconds, running on a single NVIDIA A100.
Highlights
- SAM 3.1 paired with DINOv3 creates what the author calls a "crazy combo" for visual tracking
- The system holds onto surgical tools even while a forearm sweeps over them
- It can distinguish between near-identical twin tools: one disappears for 15 seconds, and the system knows exactly which one returned
- The setup runs on a single NVIDIA A100 hosted on Hugging Face Jobs