Table 7:
Comparison of the proposed model with DeepFaceLIFT using single and multiple scales, respectively
| Model | Training Labels | MAE (VAS) |
|---|---|---|
|
| ||
| DeepFaceLift1 | VAS | 2.91 |
|
| ||
| VAS, OPI | 3.15 | |
|
| ||
| Proposed Model | VAS | 2.64 |
|
| ||
| VAS, OPI | 2.53 | |
|
| ||
| VAS, AFF, SEN | 2.48 | |
|
| ||
| VAS, AFF, SEN, OPI | 2.34 | |
For fair comparison, these results are based on our implementation of the DeepFaceLift method and ran on the same stratified train/test distribution as our model.