Table 3.
Mean absolute error (MAE) values for age estimation on UTKFace [83], for fully visible faces (denoted by 
), lower-half-visible faces (denoted by 
) and upper-half-visible faces (denoted by 
)
| Method | Train faces | Test faces | MAE |
|---|---|---|---|
| ResNet-50+PyNADA [19] |  |
 |
5.79 |
| Teacher VGG-f |  |
 |
5.63 |
| Teacher VGG-face |  |
 |
5.11 |
| Teacher ResNet-50 |  |
 |
5.27 |
| Teacher VGG-f |  |
 |
11.16 |
| Teacher VGG-face |  |
 |
13.08 |
| Teacher ResNet-50 |  |
 |
14.23 |
| Teacher VGG-f |  |
 |
9.60 |
| Teacher VGG-face |  |
 |
10.30 |
| Teacher ResNet-50 |  |
 |
11.92 |
| VGG-f |  |
 |
6.80 |
| VGG-face |  |
 |
6.15 |
| ResNet-50 |  |
 |
6.66 |
| VGG-f |  |
 |
6.36 |
| VGG-face |  |
 |
5.53 |
| ResNet-50 |  |
 |
6.44 |
| VGG-f (standard T-S) |
 + 
|
 |
6.34 |
| VGG-face (standard T-S) |
 + 
|
 |
5.40 |
| ResNet-50 (standard T-S) |
 + 
|
 |
6.35 |
| VGG-f (triplet loss T-S) |
 + 
|
 |
6.34 |
| VGG-face (triplet loss T-S) |
 + 
|
 |
5.42 |
| ResNet-50 (triplet loss T-S) |
 + 
|
 |
6.34 |
| VGG-f (triplet loss + standard T-S) |
 + 
|
 |
6.22 |
| VGG-face (triplet loss + standard T-S) |
 + 
|
 |
5.40 |
| ResNet-50 (triplet loss + standard T-S) |
 + 
|
 |
6.33 |
A state-of-the-art model [19] is included as reference