Table 3 Comparison of MNE and FR according to the experimental results for each model.

From: A rapid identification method for soft tissue markers of dentofacial deformities based on heatmap regression

Index

Network model

WFLW

300 W

Customize

Common

Challenging

Full

MNE(%)

Resnet50

7.1

8.6

15.4

10.7

Mobile-net

6.9

8.2

17.5

11.4

DeCaFa [45]

6.6

PIP-Net [18]

6.5

EfficientNet-B3 [46]

7.8

ATPN [47]

6.1

3FabRec [48]

6.2

7.3

9.7

8.3

HR-net

5.9

7.3

9.4

8.1

7.5

BHR-net

5.2

6.9

9.4

7.8

2.5

FR > 10% (%)

Resnet50

11.9

10.9

20.6

11.6

Mobile-net

14.35

10.2

29.4

17.3

DeCaFa [45]

8.9

PIP-Net [18]

8.2

EfficientNet–B3 [46]

19.7

ATPN [47]

7.4

3FabRec [48]

8.2

HR-net

7.1

8.6

16.3

12.1

7.2

BHR-net

6.9

7.8

13.7

10.6

1.4

  1. % omitted, – not counted.
  2. The data in bold are the results of the BHR-net.