Table 4.

Comparison of eXnet with benchmark networks on the FER-2013 dataset.

Model	Accuracy (%)	Params	Size (MB)
VGG [7]	71.29	14.72M	70.94
ResNet [20]	71.12	11.17M	61.18
DenseNet [47]	67.54	3.0M	59.69
DeXpression [13]	68	3.54M	57.14
Liu et al. [36]	61.74	84M	-
CNN + Support Vector Machine. [9]	71	4.92M	-
Tang [11]	69.4	7.17M	-
Shao at el. [18]	71.14	7.12M	-
eXnet (ours)	71.67	4.57M	36.49
eXnet${}^{cutout}$ (ours)	71.92	4.57M	36.49
eXnet${}^{mixup}$ (ours)	72.67	4.57M	36.49
eXnet${}^{cm}$ (ours)	73.54	4.57M	36.49