Table 3 Comparisons with other methods on the AFLW2000 and BIWI datasets. All methods are trained on the 300W-LP dataset.
Method | AFLW2000 | BIWI | ||||||
|---|---|---|---|---|---|---|---|---|
Yaw | Pitch | Roll | MAE | Yaw | Pitch | Roll | MAE | |
3DDFA38 | 5.40 | 8.53 | 8.25 | 7.39 | 36.2 | 12.3 | 8.78 | 19.1 |
FAN41 | 6.37 | 12.3 | 8.71 | 9.12 | 8.53 | 7.48 | 7.63 | 7.89 |
HopeNet24 | 6.47 | 6.56 | 5.44 | 6.16 | 4.81 | 6.61 | 3.27 | 4.90 |
QuatNet25 | 3.97 | 5.61 | 3.92 | 4.50 | 4.01 | 5.49 | 2.93 | 4.14 |
FSA-Net26 | 4.50 | 6.08 | 4.64 | 5.07 | 4.27 | 4.96 | 2.76 | 4.00 |
FDN42 | 3.78 | 5.61 | 3.88 | 4.42 | 4.52 | 4.70 | 2.56 | 3.93 |
Image2pose43 | 3.43 | 5.03 | 3.27 | 3.91 | 4.57 | 3.55 | 3.24 | 3.79 |
6DRepNet27 | 3.27 | 4.58 | 2.98 | 3.61 | 3.24 | 4.48 | 2.68 | 3.47 |
ASG Learning44 | 3.08 | 4.74 | 3.11 | 3.64 | 4.21 | 3.52 | 3.10 | 3.61 |
TokenHPE45 | 4.36 | 5.54 | 4.08 | 4.66 | 3.95 | 4.51 | 2.71 | 3.72 |
HeadDiff46 | 3.15 | 4.55 | 3.03 | 3.57 | 3.47 | 4.20 | 2.71 | 3.46 |
WQuatNet47 | 3.39 | 4.62 | 3.23 | 3.75 | 3.91 | 4.20 | 2.64 | 3.58 |
HRHPE48 | 3.33 | 4.29 | 3.20 | 3.60 | 3.38 | 4.34 | 2.52 | 3.41 |
Our | 2.94 | 4.36 | 2.69 | 3.33 | 3.83 | 3.98 | 2.32 | 3.38 |
