Table 6 Comparisons with state-of-the-art on VisDrone, UAVDT, and AU-AIR datasets (%).

From: ACFM: adaptive channel weighted fusion algorithm for improving small object detection performance in UAV traffic

Methods

Backbone

VisDrone2019

UAVDT

AUAIR

mAP↑

AP50↑

AP75↑

mAP↑

AP50↑

AP75↑

mAP↑

AP50↑

AP75↑

GFL

ResNet50

24.3

39.8

25.2

15.6

27.4

16.3

6.4

16.2

4.1

GFL + EFC

ResNet50

30.6

51.1

31.8

16.5

29.0

17.6

13.6

33.8

8.8

GFL+ACFM

ResNet50

31.2

51.8

32.2

17.3

29.6

18.1

14.0

34.5

9.3

RetinaNet

ResNet50

19.4

32.4

20.1

15.7

27.2

17.2

12.0

30.5

7.6

RetinaNet + EFC

ResNet50

24.2

40.7

24.8

15.3

26.4

16.4

16.1

38.6

11.2

RetinaNet+ACFM

ResNet50

24.8

41.3

25.2

15.9

27.8

17.1

16.5

39.7

11.6

FSAF

ResNet18

18.7

34.0

17.7

15.3

27.3

15.7

12.3

31.6

6.9

FSAF + EFC

ResNet18

25.5

47.2

24.7

15.9

28.4

16.4

14.8

36.1

9.1

FSAF+ACFM

ResNet18

26.4

48.3

25.3

16.9

29.8

17.8

15.3

36.6

10.2

FCOS

ResNet50

17.4

30.7

16.8

14.8

27.5

15.1

10.3

25.9

5.8

FCOS + EFC

ResNet50

18.6

32.6

18.7

15.3

27.9

15.5

11.0

28.7

6.4

FCOS+ACFM

ResNet50

19.3

33.2

19.1

16.2

29.0

16.9

13.0

33.2

7.1