YOLO-style Detectors

Best for: Real-time object detection Aliases: YOLO, You Only Look Once, Single-Stage Detector

How it works

A single fully-convolutional network divides the image into an $S\times S$ grid; each grid cell directly regresses $B$ bounding boxes with coordinates $(x,y,w,h)$, an objectness score $\text{conf}=P(\text{obj})\cdot\text{IoU}$, and class probabilities, all in one forward pass with no region-proposal stage. Training minimises a multi-part loss over localization, confidence, and classification, and non-max suppression removes duplicate boxes at inference. The single-shot design is what enables real-time detection.

When to use

Real-time object detection on edge or streaming devices where the speed-accuracy tradeoff matters.

Watch out

Lower accuracy than two-stage detectors on small or dense objects; needs careful anchor and NMS tuning.

Common fields

Surveillance · robotics · retail · autonomous systems