yolov5

The model detects and localizes multiple objects in images simultaneously, producing bounding boxes with class labels and confidence scores in a single forward pass through the networkthe architecture that made YOLO (You Only Look Once) the reference standard for real-time detection.

YOLOv5 is available in multiple size variants (nano through extra-large) calibrated for different speed-accuracy tradeoffs, with the nano and small variants running at hundreds of frames per second on GPU hardware and the larger variants providing state-of-the-art accuracy on the COCO benchmark.

The Ultralytics implementation includes comprehensive training infrastructure: data augmentation (mosaic, mixup, CopyPaste), automatic learning rate scheduling, model export to ONNX, TensorRT, CoreML, and TFLite formats, and a validation pipeline with standard COCO metrics.

Computer vision engineers building surveillance systems, autonomous vehicle perception stacks, robotics vision systems, retail analytics, medical imaging pipelines, and quality inspection applications use YOLOv5 as the detection backbone.

Its combination of ease of use (training a custom model requires a few lines of Python), active maintenance, and mature export tooling for edge deployment makes it a reliable production choice.

Though superseded in benchmark accuracy by YOLOv8 and subsequent versions from Ultralytics, YOLOv5's stability and the volume of existing deployments make it a continuing reference point in object detection.