ultralytics
**Ultralytics YOLO** is the **most widely used real-time object detection framework, providing the YOLOv5 and YOLOv8 model families with a "zero to hero" developer experience** — enabling training of state-of-the-art detection, segmentation, pose estimation, and classification models on custom datasets with as few as 3 lines of Python code, automatic export to every major deployment format (ONNX, CoreML, TFLite, TensorRT), and real-time inference on webcams, edge devices, and production servers.
**What Is Ultralytics YOLO?**
- **Definition**: A Python framework by Ultralytics (founded by Glenn Jocher) that implements the YOLO (You Only Look Once) family of single-stage object detection models — where the entire image is processed in one forward pass through the network, predicting bounding boxes and class probabilities simultaneously for real-time speed.
- **YOLOv8**: The latest generation supporting four task types in a unified architecture — object detection (bounding boxes), instance segmentation (pixel masks), pose estimation (keypoint skeletons), and image classification — all trained and deployed through the same CLI and Python API.
- **Speed/Accuracy Balance**: YOLO models are the default choice for real-time applications — YOLOv8n (nano) runs at 300+ FPS on a modern GPU for edge deployment, while YOLOv8x (extra-large) achieves 53.9% mAP on COCO for maximum accuracy.
- **Ecosystem**: Built-in data augmentation (mosaic, mixup, copy-paste), automatic hyperparameter tuning, Weights & Biases / MLflow / Comet integration, and one-command export to 11+ deployment formats.
**YOLOv8 Model Variants**
| Model | Params | mAP (COCO) | Speed (A100) | Use Case |
|-------|--------|-----------|-------------|----------|
| YOLOv8n | 3.2M | 37.3% | 0.99ms | Edge, mobile, IoT |
| YOLOv8s | 11.2M | 44.9% | 1.20ms | Balanced speed/accuracy |
| YOLOv8m | 25.9M | 50.2% | 1.83ms | General production |
| YOLOv8l | 43.7M | 52.9% | 2.39ms | High accuracy |
| YOLOv8x | 68.2M | 53.9% | 3.53ms | Maximum accuracy |
**Key Features**
- **3-Line Training**: `from ultralytics import YOLO; model = YOLO("yolov8n.pt"); model.train(data="coco128.yaml", epochs=100)` — complete training pipeline with augmentation, validation, and checkpointing in three lines.
- **Multi-Task Architecture**: The same YOLOv8 backbone supports detection (`yolov8n.pt`), segmentation (`yolov8n-seg.pt`), pose (`yolov8n-pose.pt`), and classification (`yolov8n-cls.pt`) — switch tasks by changing the model file.
- **Export to 11+ Formats**: `model.export(format="onnx")` converts to ONNX, TensorRT, CoreML, TFLite, OpenVINO, PaddlePaddle, NCNN, and more — one command for any deployment target.
- **Real-Time Inference**: `model.predict(source="webcam")` runs real-time detection on webcam, video files, RTSP streams, or image directories — with built-in visualization and tracking (ByteTrack, BoT-SORT).
- **Ultralytics HUB**: Cloud platform for dataset management, model training, and deployment — train YOLOv8 models without local GPU resources.
**YOLO vs Other Detection Frameworks**
| Framework | Ease of Use | Speed | Accuracy | Research Flexibility |
|-----------|------------|-------|----------|---------------------|
| Ultralytics YOLO | Excellent | Fastest | Very good | Moderate |
| MMDetection | Moderate | Good | Excellent | Excellent |
| Detectron2 | Moderate | Good | Excellent | Excellent |
| TF Object Detection | Complex | Good | Good | Good |
| DETR (Transformers) | Moderate | Slower | Excellent | Excellent |
**Ultralytics YOLO is the real-time object detection framework that makes production computer vision accessible to every developer** — combining state-of-the-art accuracy with unmatched ease of use, multi-task support, and universal export capabilities that take a custom detection model from training to edge deployment in minutes rather than weeks.