You Only Look Once (YOLO)

Redmon et al. (2015)

Read paper

Why It Matters

Object detection as single regression at 45 FPS. Made real-time detection practical for robotics, surveillance, autonomous systems.

Key Ideas

1. Treat object detection as one direct regression problem from image pixels to bounding boxes and class probabilities instead of a proposal stage followed by classification.
2. Predict all detections in a single forward pass, making real-time object detection practical.
3. Accept weaker localization and small-object recall in exchange for simplicity and speed.
4. Establish the one-stage detector pattern that later real-time detectors refined rather than replaced.

Notes

• YOLO changed the culture of detection by making speed a primary design goal.
• Later YOLO variants mostly improve the same basic idea with better training, scaling, and box prediction.