Table 4 Performance comparison of different methods on the LLVIP dataset.

From: Cross-modal edge-enhanced detector for UAV-based multispectral object detection

Methods

Backbone

mAP@50

mAP@[0.5:0.95]

Modality

SSD54

VGG16

82.6

39.8

 

RetinaNet40

ResNet50

88.0

42.8

 

CascadeR-CNN55

ResNet50

88.3

47.0

RGB

Faster R-CNN41

ResNet50

87.0

45.1

 

DDQ-DETR56

ResNet50

86.1

46.7

 

SSD54

VGG16

90.2

53.3

 

RetinaNet40

ResNet50

94.8

55.1

 

Cascade R-CNN55

ResNet50

95.0

56.8

IR

Faster R-CNN41

ResNet50

94.6

54.5

 

DDQ-DETR56

ResNet50

93.9

58.6

 

Halfway Fusion45

VGG16

91.4

55.1

 

GAFF57

ResNet18

94.0

55.8

 

ProbEn22

ResNet50

93.4

51.5

RGB + IR

CSAA58

ResNet50

94.3

59.2

RSDet59

ResNet50

95.8

61.3

 

CMEE-Det

CSPDarknet53

97.0

64.7

Â