Table 3.
Comparison of performance with previous FER methods on the test set of the RAF-DB dataset after incorporating random occlusion.
| Method | Acc.(%) |
|---|---|
| Baseline | 81.62 |
| SCN (Wang et al., 2020) | 85.78 |
| MA-Net (Zhao et al., 2021) | 86.23 |
| FFRA-Net (Ours) | 86.43 |
The bold values are outcomes from model runs described in this paper.