Table 2 Outdoor pose estimation on MegaDepth45.

From: Swin-transformer for weak feature matching

Category

Method

Pose estimation AUC

P

@\({5}^{\circ }\)

@\({10}^{\circ }\)

@\({20}^{\circ }\)

Detector-based

SIFT7+SuperGlue19CVPR’20

35.32

56.12

73.68

79.85

ORB9+GMS46CVPR’17

35.05

54.18

70.92

75.79

Sp47+PointCN48CVPR’18

40.34

57.05

74.81

79.16

SP47+SuperGlue19CVPR’20

42.18

61.16

75.96

85.19

SP47+OminiGlue54CVPR’24

47.40

65.00

77.80

Detector-free

LoFTR17CVPR’21

52.80

69.19

81.18

94.80

MatchFormer18ACCV’22

48.74

65.83

78.81

97.55

Aspanformer49ECCV’22

55.30

71.50

83.10

97.48

OAMatcher50Pattern Recogn’24

56.56

72.34

83.61

MR-Matche51IEEE Trans’24

55.96

72.15

83.45

PDC-Net+52TPAMI’23

43.10

61.90

76.10

PCFs53TPAMI’23

49.81

65.40

78.98

SwinMatcher

56.21

73.90

83.70

97.86