github.com/ultralytics/yolov5 ↗

About ultralytics/yolov5

YOLOv5 is a state-of-the-art computer vision model built on PyTorch that specializes in object detection, image segmentation, and image classification tasks. Developed by Ultralytics, it is designed for ease of use, speed, and accuracy, incorporating insights from extensive research and development. The project provides pretrained models of varying sizes (nano through extra-large) that balance performance and computational requirements, with detailed performance metrics on the COCO dataset.

The repository enables users to perform inference on various sources using scripts like detect.py, train custom models on their own datasets, and export models to multiple deployment formats including ONNX, CoreML, TFLite, and TensorRT. This multi-format export capability makes YOLOv5 adaptable for deployment across different platforms and hardware, from edge devices to cloud infrastructure. The project includes comprehensive documentation, tutorials covering topics like multi-GPU training, hyperparameter evolution, and model optimization through pruning and quantization.

YOLOv5 serves a broad audience from researchers and computer vision practitioners to production teams building real-world applications. The project maintains an active community through Discord and provides integrations with popular platforms like Weights & Biases, Comet ML, and Roboflow for experiment tracking and model management. An official Ultralytics Platform offers a no-code interface for labeling datasets, training models, and deploying them to production environments.