Table 3 Quantitative Analysis of Ablation Exam Results.

From: A monocular endoscopic image depth estimation method based on a window-adaptive asymmetric dual-branch Siamese network

Network setting

HyperKvasir

EAD

D1 − all(%)↓

 

3px(%) ↑

 

D1 − all(%) ↓

 

3px(%) ↑

 

G-B

L-B

N

NOC

ALL

NOC

ALL

NOC

ALL

NOC

ALL

–

✓

0

1.53

1.44

56.5

34.3

1.46

1.39

43.9

42.1

✓

–

0

1.46

1.36

52.1

58.2

1.44

1.35

56.2

47.8

✓

✓

0

1.31

1.3

68.2

64.5

1.38

1.3

69.7

63.7

✓

✓

1

1.16

1.29

74.4

72.4

1.32

1.24

70.5

69.3

✓

✓

2

0.87

0.31

84.1

82.9

0.4

0.31

81

80.2

✓

✓

3

0.46

0.23

89.6

87

0.32

0.26

88.2

85.3

  1. NOC (Non-occluded) refers to evaluations conducted only in non-occluded areas, meaning pixels that are not occluded. ALL indicates evaluations across the entire image, including both occluded and non-occluded areas. This metric is generally more challenging because depth estimation in occluded areas is more difficult.
Back to article page