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

YOLO takes a different approach to object detection than prior systems, which repurpose classifiers to perform detection. Instead, it treats detection as a single regression problem, using one neural network to predict bounding boxes and class probabilities directly from a full image in a single evaluation. Because the entire detection pipeline is one network, it can be optimized end-to-end directly on detection performance.

The resulting architecture is extremely fast: the base YOLO model runs at 45 frames per second, and a smaller Fast YOLO variant reaches 155 fps while still doubling the mean average precision of other real-time detectors. YOLO trades off some localization accuracy for fewer background false positives compared to systems like DPM and R-CNN, and it was shown to generalize better than those methods when applied to non-natural image domains such as artwork, indicating it learns more general object representations.