Table 3 Experimental results of different object detection models on DOTA test set.
From: A real-time and efficient detector for small object in UAV aerial images
Model | Precision/% | Recall/% | mAP50/% | mAP50:95/% | Parameters/10\(^6\) | FPS | GFLOPS |
|---|---|---|---|---|---|---|---|
YOLOv5s16 | 41.5 | 33.3 | 31.3 | 17 | 7.04 | 180 | 7.04 |
YOLOv8s36 | 45.9 | 32.9 | 32.9 | 18.1 | 3 | 182 | 3 |
YOLOv9t21 | 40.4 | 34.6 | 30.8 | 16.7 | 2.6 | 135 | 10.7 |
YOLO10s17 | 39.7 | 34.3 | 32.5 | 18.1 | 8.04 | 153 | 8.2 |
YOLO11n36 | 46.5 | 32 | 31.7 | 16.6 | 2.58 | 196 | 2.58 |
RT-DETR20 | 48.9 | 40.3 | 38 | 20 | 19.8 | 97 | 57 |
RT-UAV-SOD (ours) | 53.2 | 42.4 | 39.1 | 22.6 | 17.7 | 83 | 60 |
