Skip to content

SAMURAI

Zero-shot visual tracking with motion memory

Visit official site

What SAMURAI is

The model is described as a visual object tracker built on Segment Anything Model 2 for fast-moving or self-occluding objects.

What you can do with it

  • Track objects without fine-tuning
  • Select masks with motion cues
  • Run tracking in real time

Application scenarios

  • Follow athletes in a match
  • Track people in surveillance
  • Track moving objects for driving

Pros & cons

✓ What we like

  • No fine-tuning is needed to generalise
  • LaSOT-ext AUC is stated to rise by 7.1 percent

! What to watch out for

  • Tracking builds on the SAM 2 model
  • Mask selection relies on motion cues

FAQ

Does it need retraining?

The material states no retraining or fine-tuning is needed.

What gain is listed?

The material states a 7.1 percent AUC gain on LaSOT-ext and 3.5 percent AO on GOT-10k.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review