Table 1.

Classification accuracies on the FER 2013 dataset.

Method	Accuracy Rate
Unsupervised Domain Adaptation [47]	65.3%
Bag of Words [48]	67.4%
VGG+SVM [49]	66.31%
GoogleNet [50]	65.2%
FER on SoC [51]	66%
Mollahosseini et al. [9]	66.4%
The proposed algorithm	70.02%
Aff-Wild2 (VGG backbone) [52]	75%