You Only Look Once: Unified, Real-Time Object Detection (YOLO)

Reframed object detection from a multi-stage pipeline into a single regression problem from full image pixels to bounding box coordinates and class probabilities.

The undisputed global industry standard family (YOLOv8, YOLO11) for real-time video surveillance, robotics, drones, and autonomous vehicles.

Assumptions

• Grid cells can simultaneously predict bounding boxes, confidence scores, and class conditional probabilities in a single forward pass

Limitations

• Initial version struggled with small object detection and crowded clusters of items in a single grid cell