ultralytics/ultralytics

Ultralytics (ultralytics/ultralytics)

Ultralytics is an open-source project and software distribution that packages the company's YOLO models and supporting tooling into an easy-to-use Python library and command-line interface. The repository is focused on providing a complete computer-vision stack: ready-made SOTA YOLO models, training and evaluation utilities, pretrained weights, export tools for common deployment formats, and integrations with popular ML platforms.

Key features

• Comprehensive YOLO model family: includes lightweight to large variants suited for different accuracy/latency trade-offs.
• Multi-task support: detection, instance segmentation, pose estimation, classification and tracking are supported in a single codebase.
• Easy installation and use: pip-installable package and a yolo CLI plus a pythonic API (from ultralytics import YOLO) for scripting and interactive use.
• Pretrained weights & model hub: models are distributed as pretrained weights for immediate inference and fine-tuning.
• Export & deployment: utilities to export models to ONNX and other formats to accelerate inference on CPU/GPU/edge devices.
• Integrations: built-in compatibility with common ML tooling (experiment trackers, dataset tools, and inference/optimization libraries).
• Active community & docs: detailed documentation, examples (Colab / Kaggle / Binder), community forums, Discord and issue tracker for support and contributions.

Typical workflows

1. Install the package with pip or use the repository directly for development.
2. Use pretrained models for inference (yolo predict) or load models in Python for programmatic use.
3. Prepare datasets in supported formats and run training with simple CLI or Python calls.
4. Validate and export models for deployment, or integrate with experiment tracking and optimization tools.

Licensing & commercial use

The repository is published under AGPL-3.0 for the open-source distribution and Ultralytics additionally offers an enterprise license for commercial deployments that require different licensing terms.

Who should use it

• Researchers who want reference implementations and reproducible YOLO training/evaluation pipelines.
• Engineers building production CV systems who need pretrained models, export tools and deployment-ready utilities.
• Learners and students who want a clear, well-documented path from training to inference with real-world models.

Quickstart example (Python)

from ultralytics import YOLO
model = YOLO('yolo26n.pt')
results = model('path/to/image.jpg')
results[0].show()

Overall, ultralytics/ultralytics packages decades of YOLO research and engineering into a practical, production-minded toolkit for modern computer vision projects.