Yolov5
YOLOv5 is a family of compound-scaled object detection models trained on the COCO dataset. It is implemented in PyTorch and provides a complete training, validation, and inference pipeline. The model architecture uses a modified CSPDarknet53 backbone with a PANet neck and YOLO detection head, supporting anchor-based prediction across multiple scales. YOLOv5 ships with several pre-trained variants—nano, small, medium, large, and xlarge—that trade speed for accuracy. The repository includes utilities for automatic data augmentation (Mosaic, MixUp), hyperparameter evolution, auto-anchor selection, and export to deployment formats like ONNX, CoreML, TFLite, TensorRT, and OpenVINO. A classification module supports image classification tasks, and the codebase is designed for easy extension and customization. YOLOv5 has been widely adopted due to its balance of speed and accuracy, active community maintenance, and integration with the Ultralytics ecosystem. It can be used from the command line or as a Python library, and Docker containers are provided for reproducible execution. The project is under continuous updates with frequent commits addressing bugs, dependency bumps, and compatibility improvements.
- Category
- AI/ML
- Pricing
- Open Source
Problem solved
YOLOv5 solves the problem of real-time object detection with high accuracy and low computational cost. Developers and researchers need a state-of-the-art detection system that can be deployed on edge devices like smartphones, embedded platforms, and browsers. YOLOv5 provides a single-stage detector that processes images and video streams quickly while maintaining competitive mAP scores, eliminating the need for complex two-stage detectors. Its extensive export capabilities further simplify deployment across diverse hardware.
Use cases
- Real-time object detection in live video feeds (surveillance, smart cities)
- Inventory counting and product identification in retail warehouses
- Automatic license plate recognition
- Animal species detection and counting in wildlife monitoring
- Defect detection on manufacturing assembly lines
- Face mask detection in public health applications
Audience
- Developers
- ML Engineers
- Data Scientists
- Researchers
Pros
- Extremely fast inference with real-time performance
- High accuracy on standard benchmarks (MS COCO)
- Easy export to multiple deployment formats (ONNX, TFLite, etc.)
- Large community and active maintenance
- Multiple model sizes allow flexible speed/accuracy tradeoffs
Cons
- AGPL-3.0 license can be restrictive for commercial, closed-source projects
- Limited to object detection and classification; no built-in instance segmentation
Official website: https://github.com/ultralytics/yolov5
Canonical LambdaBase page: https://www.lambdabase.com/tools/yolov5