Table 2 Ensemble landmarking and PC results.

From: Ensemble landmarking of 3D facial surface scans

Landmark

[Benchmark]

SoWR 15

Mean 15

PC

[Bagging]

[Stacking]

1

2.4

(3.1)

2.4

3.1

4.4

3.7

1.8

(1.3)

2

1.7

(0.8)

2.4

2.8

2.8

2.4

2.1

(1.4)

3

1.8

(1.0)

1.9

1.9

2.6

1.8

1.6

(1.0)

4

2.3

(1.7)

1.7

2.1

2.7

1.4

1.5

(0.8)

5

2.4

(1.9)

2.9

2.4

2.8

2.2

2.0

(1.6)

6

6.5

(4.3)

4.7

3.3

4.9

5.8

3.0

(2.0)

7

2.1

(1.3)

1.7

1.9

14.5

3.0

1.4

(0.6)

8

1.5

(1.3)

2.0

2.2

3.2

1.5

1.9

(1.0)

9

1.8

(1.2)

2.7

2.7

4.4

2.8

2.3

(1.9)

10

2.2

(1.2)

1.7

2.4

3.9

3.0

1.6

(0.9)

11

1.7

(1.9)

1.6

1.6

3.1

1.6

1.3

(0.8)

12

1.1

(0.7)

1.9

1.7

12.2

1.6

1.3

(0.8)

13

1.4

(0.9)

2.0

1.6

13.1

1.5

1.5

(0.7)

14

1.2

(0.6)

2.2

1.7

10.5

2.2

1.5

(0.9)

15

1.5

(0.8)

1.6

1.5

13.5

1.5

1.3

(0.7)

16

1.2

(0.7)

1.7

1.9

14.1

2.0

1.7

(1.0)

17

1.2

(2.3)

1.8

2.9

14.3

2.6

1.5

(0.9)

18

2.0

(1.4)

1.4

4.4

18.3

2.8

1.4

(0.9)

19

2.5

(3.4)

3.1

3.4

15.8

3.6

2.1

(2.3)

20

1.9

(3.4)

1.8

4.1

18.4

3.0

1.8

(1.7)

21

3.3

(5.4)

3.1

6.1

2.1

10.0

2.0

(1.2)

mn

2.1

 

2.2

2.6

8.6

2.9

1.7

 

sd

1.3

0.8

1.1

6.0

1.9

0.4

  1. Results are reported in Euclidean distance to manual training data in mm. Benchmark represent results from the previous version of our algorithm15. Clarification of terms: SoWR 15 = based on intermediate Summation of Wavelet Responses of 15 landmarkers. Mean 15 = mean of final coordinates of 15 landmarkers. PC = results obtained by our principal component method. Distances <2 mm are underlined, distances >4 mm are in italics. Standard deviations are shown in parentheses.
Back to article page