Vision · 2016– · real-time detection
YOLO
You Only Look Once: object detection formulated as a single regression problem over a grid, enabling real-time detection in one network pass.
ModLensVision
Interactive Diagram
Focus the lens
Click any component to read what it does. Signal direction follows the edges.
inputFull Image. The whole scene in one pass — no region proposals.
Core idea
Detection as direct prediction: divide the image into cells, and let each cell regress boxes and classes simultaneously. Speed comes from doing everything once.
Why it exists
Two-stage detectors were accurate but slow; many applications (video, robotics, edge) need decisions per frame.
Data Flow
What moves through the system
- 01The full image enters a single backbone.
- 02A neck fuses multi-scale features.
- 03Dense heads predict boxes, objectness and classes per cell.
- 04Non-maximum suppression yields final detections.
Strengths
- + Real-time throughput
- + Single-pass simplicity
- + Strong modern variants at every size
Limitations
- − Historically weaker on small objects
- − Dense prediction needs careful label assignment
Applications
- · Video analytics
- · Robotics
- · Traffic and safety systems
- · Edge devices