Table 1.
The performance of several object detection models
| Models | Feature extraction network | Object detection | mAP | LAMR | F1 | Precision | Recall |
|---|---|---|---|---|---|---|---|
| MobileNet-SSD | MobileNet | SSD | 85.73% | 2.6×10-1 | 0.76 | 93.55% | 64.44% |
| VGG16-SSD | VGG16 | SSD | 77.99% | 3.5×10-1 | 0.74 | 77.78% | 70.00% |
| YOLO-V3 | Darknet53 | YOLO-V3 | 74.90% | 3.1×10-1 | 0.75 | 100.00% | 60.00% |
| YOLO-V4 | CSPDarknet53 | YOLO-V4 | 85.36% | 2.4×10-1 | 0.88 | 97.22% | 79.55% |
| YOLO-V4-tiny | CSPDarknet53 | YOLO-V4-tiny | 81.00% | 2.9×10-1 | 0.75 | 100.00% | 60.00% |
| RFB-SSD | VGG-RFB | SSD | 96.36% | 2.9×10-6 | 0.90 | 90.00% | 90.00% |
| Retinanet | ResNet50 | Retinanet | 89.28% | 2.9×10-3 | 0.86 | 81.82% | 90.00% |
| M2det | VGG16 | M2det | 71.54% | 3.0×10-1 | 0.82 | 100.00% | 70.00% |
| CenterNet | ResNet50 | CenterNet | 73.47% | 3.6×10-1 | 0.63 | 95.45% | 46.67% |
| EfficientDet D0 | EfficientNetD0 | EfficientDet | 94.08% | 1.2×10-1 | 0.85 | 90.00% | 80.00% |
| EfficientDet D1 | EfficientNetD1 | EfficientDet | 94.34% | 1.2×10-1 | 0.86 | 90.24% | 82.22% |
| Faster RCNN | ResNet50 | Faster RCNN | 88.18% | 2.1×10-1 | 0.72 | 60.00% | 90.00% |