Display Format
Bounding Box Coordinates
XYXY (x₁, y₁, x₂, y₂)
No box drawn
XYWH (x, y, width, height)
No box drawn
Normalized XYWH (0-1)
No box drawn
Center XYWH (cx, cy, w, h)
No box drawn
| Model | Summary | Bounding Box Format |
|---|---|---|
| YOLO | Real-time object detection system. Training format uses normalized coordinates for better generalization across different image sizes. | Normalized XYWH (x, y, w, h) in range [0,1] |
| Faster R-CNN | Region proposal network based detection. Typically uses absolute coordinates during inference, but may use normalized coordinates during training. | XYXY (x₁, y₁, x₂, y₂) or normalized coordinates |
| SSD | Single Shot MultiBox Detector uses normalized coordinates for anchor boxes and predictions to handle multiple scales efficiently. | Normalized XYWH (x, y, w, h) in range [0,1] |
| RetinaNet | Dense object detector with focal loss. Uses normalized coordinates for anchor boxes and predictions. | Normalized XYWH (x, y, w, h) in range [0,1] |
| CornerNet | Anchor-free detector that predicts keypoint heatmaps. Uses normalized coordinates for better scale invariance. | Normalized Corners (x₁, y₁, x₂, y₂) in range [0,1] |