Table 7 Comparative experimental results on VisDrone 2019 test set.
Method | AP (%) | \(\hbox {AP}_{50}\) (%) | \(\hbox {AP}_{small}\) (%) | Params | GFLOPs | FPS |
|---|---|---|---|---|---|---|
Faster-RCNN-R50 36 | 19.4 | 32.9 | 9.5 | 41.39M | 208G | 53.3 |
Cascade-RCNN-R50 37 | 19.7 | 32.6 | 9.9 | 69.29M | 236G | 44.1 |
RetinaNet-R50 38 | 16.4 | 27.6 | 6.0 | 36.52M | 210G | 56.7 |
YOLOv8n 39 | 14.4 | 25.9 | 5.9 | 3.0M | 8.1G | 238.1 |
YOLOv8s 39 | 17.3 | 30.7 | 7.8 | 11.13M | 28.5G | 158.7 |
YOLOv8m 39 | 19.0 | 33.2 | 9.0 | 25.85M | 78.7G | 135.5 |
YOLOv10n 40 | 14.2 | 26.1 | 6.3 | 2.28M | 6.5G | 244.4 |
YOLOv10s 40 | 17.9 | 32.3 | 8.6 | 7.22M | 21.4G | 172.4 |
YOLOv10m 40 | 19.5 | 34.5 | 9.7 | 15.32M | 58.9G | 105.4 |
YOLOv11n 41 | 14.2 | 25.8 | 5.8 | 2.59M | 6.3G | 250.0 |
YOLOv11s 41 | 17.6 | 31.3 | 8.0 | 9.42M | 21.3G | 175.4 |
YOLOv11m 41 | 20.3 | 35.0 | 9.8 | 20.40M | 67.7G | 117.1 |
RT-DETR-R18 28 | 20.8 | 36.3 | 15.7 | 19.86M | 57G | 78.1 |
RT-DETR-R34 28 | 20.8 | 36.2 | 16.1 | 31.44M | 90.6G | 68.3 |
RT-DETR-R50 28 | 21.6 | 37.8 | 16.9 | 42.94M | 134.8G | 58.1 |
Proposed Method | 23.1 | 39.8 | 18.8 | 14.84M | 65.7G | 76.9 |
