Table 4.
Comparison of different models on VisDrone-test-dev set.
| Model | Precision | Recall | mAP@.5% | mAP@.5:.95% |
|---|---|---|---|---|
| SSD51230 | 11.0 | 40.5 | 23.9 | – |
| FPN31 | 27.3 | 39.7 | 29.2 | – |
| RetinaNet23 | 13.8 | 29.9 | 21.2 | – |
| YOLOx-s32 | 24.6 | 44.6 | 33.8 | 20.2 |
| YOLOx-l | 35.4 | 44.4 | 37.1 | 21.1 |
| YOLOv311 | 45.9 | 34.8 | 32.3 | 18.3 |
| YOLOv3-spp | 49.4 | 33.7 | 32.4 | 18.1 |
| YOLOv5-s | 43.8 | 34.3 | 31.7 | 17.6 |
| YOLOv5-l | 31.4 | 46.2 | 33.9 | 19.2 |
| YOLOv7tiny33 | 41.2 | 33.7 | 28.8 | 14.5 |
| ImproveYOLOv535 | 36.9 | 49.6 | 38.5 | 21.8 |
| EL-YOLO-s34 | 54.1 | 44.5 | – | 21.4 |
| FocusDet(ours) | 46.0 | 35.4 | 40.6 | 23.9 |
The bolded performance is the best one.