Table 16 Comparison of state-of-the-art UAV image detection models on the VisDrone 2019 test set.
From: End to end polysemantic cooperative mixed task trainer for UAV target detection
Method | FPS | AP_s | mAP0.5 | pedestrian | people | bicycle | car | van | trunk | tricycle | awning-tricycle | bus | motor |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Faster R-CNN20 | 19 | 15.1 | 36.2 | 37.4 | 18.4 | 13.2 | 71.8 | 42.4 | 42.8 | 19.7 | 18.2 | 58.1 | 34.3 |
YOLOv5l69 | 16 | 11.9 | 37.4 | 44.3 | 36.7 | 15.7 | 73.8 | 39.1 | 36.3 | 22.7 | 12.1 | 50.2 | 42.8 |
CGMDet70 | 13 | 30.2 | 50.7 | 58.9 | 51.1 | 24.7 | 86.1 | 52.9 | 47.6 | 38.1 | 20.2 | 66.1 | 61.7 |
YOLOv8l71 | 17 | 29.7 | 44.1 | 46.7 | 37.3 | 18.2 | 82.1 | 48.8 | 42.7 | 34.3 | 18.1 | 62.7 | 50.1 |
VitDet46 | 14 | 10.1 | 34.5 | 27.19 | 18.33 | 14.79 | 55.24 | 39.77 | 40.13 | 27.79 | 35.05 | 44.99 | 41.73 |
TPH-YOLOv5-ensemble72 | 18 | 12.6 | 36.83 | 29.13 | 16.77 | 15.78 | 67.46 | 49.75 | 44.32 | 26.59 | 25.15 | 62.15 | 31.2 |
MFEFNet73 | 14 | 30.7 | 51.9 | 59.9 | 51.2 | 25.9 | 85.8 | 55.3 | 49.4 | 40.3 | 23.2 | 65.1 | 63.1 |
Pc-DETR(Ours) | 19 | 36.9 | 54.9 | 61.66 | 53.27 | 24.77 | 87.62 | 56.11 | 50.13 | 41.22 | 43.98 | 65.91 | 64.33 |
