yangchris11/samurai

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

7112 stars · 502 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

SAMURAI is the official implementation of a zero-shot visual object tracker built on top of Segment Anything Model 2 (SAM 2), using a motion-aware memory selection mechanism to improve tracking without retraining. It runs in real time and achieves state-of-the-art zero-shot results on benchmarks like LaSOT, LaSOT-ext, GOT-10k, NFS, and OTB-2015.

Use cases

• track and segment an object across a video without training
• zero-shot visual object tracking on benchmark datasets
• segment fast-moving or occluding objects in crowded scenes
• run SAM 2 with motion-aware memory for better mask selection
• evaluate a tracker on LaSOT or GOT-10k
• multi-GPU batch inference on videos

When to choose

• you need training-free, zero-shot object tracking with segmentation masks
• you want strong tracking of fast-moving or self-occluding objects
• you are benchmarking trackers on standard VOT datasets
• you already use SAM 2 and need better temporal consistency

When to avoid

• you need a lightweight CPU-only tracker
• you require a production-ready API or GUI rather than research scripts
• you need fully supervised fine-tuned tracking accuracy on a specific domain
• you need real-time tracking on edge devices with limited GPU resources

Facets

library · maturity active

computer-vision machine-learning image-processing video-processing computer-vision machine-learning deep-learning artificial-intelligence python visual-object-tracking segment-anything sam2 zero-shot motion-aware-memory video-segmentation research-code gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="yangchris11/samurai")