YOLO is first designed by Joseph Redmon and it has developed from YOLOv1 to YOLOv3.

YOLO uses a single CNN model to achieve end-to-end object detection.

First, it resizes the whole image to S x S small blocks to find the center point and each point will predict bounding box and confidence score.

YOLO also uses IOU between predict and ground truth to show the accuracy of the bounding box. The confidence score shows the probability that this block shows the target instead of background. NMS is also been used to solve overfitting problem.