SAMURAI
SAM 2-based video object tracking model — no training required
A SAM 2-based video object tracking model that requires no training.
Image generation and computer vision — detection, editing, 3D.
Key features
- Adapts Segment Anything Model (SAM 2) for zero-shot visual tracking with motion-aware
- Uses SAM 2.1 weights directly; no additional training required
- Kalman filter estimates object state for robust motion-aware tracking
- Demo script supports inference on custom videos with bounding-box prompts
- State-of-the-art results on LaSOT, GOT-10k, TrackingNet, OTB visual tracking benchmarks
Good for
- Zero-shot visual object tracking in video without training
- Researchers benchmarking visual object tracking and video segmentation
Stack & tags
PythonPyTorchvisual-trackingsamzero-shotvideomotion-awareuwcomputer-vision
Quick start
cd sam2
pip install -e .
python scripts/demo.py --video_path <your_video.mp4> --txt_path <path_to_first_frame_bbox.txt>From the project README on GitHub.
Similar open-source projects
FLUX.2 klein
★ 25,996Black Forest Labs' sub-second image generation — generate and edit in <0.5s
DiffSynth-Studio
★ 13,200The framework behind Qwen-Image-i2L — the first "Image→LoRA" visual generation system