Table 5 The precision of the rough (first DiffusionNet) and refined (second DiffusionNet) is determined by computing the Euclidean distance between the DiffusionNet-predicted and manually annotated landmarks and is stated in millimeters ± standard deviation.

From: Fully automated landmarking and facial segmentation on 3D photographs

 

Exocanthion

Endocanthion

Nasion

Nose tip

Alare

Cheilion

Right

Left

Right

Left

  

Right

Left

Right

Left

Rough predictions

2.94 ± 2.38

2.86 ± 1.81

2.76 ± 2.40

2.83 ± 2.56

1.69 ± 1.05

1.58 ± 0.89

2.41 ± 1.98

2.52 ± 1.92

3.48 ± 3.67

3.51 ± 2.89

Refined predictions

2.25 ± 1.23

2.03 ± 1.27

1.37 ± 0.86

1.48 ± 1.00

1.48 ± 1.02

1.14 ± 0.73

1.79 ± 1.07

1.75 ± 1.11

1.71 ± 1.26

1.88 ± 1.34

Back to article page