Table 3 Summary of performance metrics for different deep learning architectures applied to the test set of the benchmark.

DL Architecture	Scale	Length	Use DA	ACC	F1	AUC	EER	ACE	Recall	Precision	Specificity	FP	FN	TP	TN
AffectNet	None	1000	False	87.67	88.18	94.16	12.00	12.33	92.00	84.66	83.33	25	12	138	125
AffectNet + FW	None	207	True	91.00	91.26	95.09	10.00	9.00	94.00	88.68	88.00	18	9	141	132
AlexNet	None	1000	False	79.33	80.13	85.56	22.00	20.67	83.33	77.16	75.33	37	25	125	113
AlexNet + FW	Standard	39	True	81.67	82.54	87.66	20.00	18.33	86.67	78.79	76.67	35	20	130	115
ResNet-50	None	1000	False	75.67	76.68	82.45	24.00	24.33	80.00	73.62	71.33	43	30	120	107
ResNet-50 + FW	Standard	30	False	77.00	77.67	79.29	24.67	23.00	80.00	75.47	74.00	39	30	120	111
VGG16	None	1000	False	73.00	73.27	80.27	27.33	27.00	74.00	72.55	72.00	42	39	111	108
VGG16 + FW	None	33	True	74.00	75.32	77.85	27.33	26.00	79.33	71.69	68.67	47	31	119	103
VGG19	None	1000	False	72.00	73.08	78.36	28.67	28.00	76.00	70.37	68.00	48	36	114	102
VGG19 + FW	Robust	40	True	74.00	75.62	79.07	29.33	26.00	80.67	71.18	67.33	49	29	121	101
ViT	None	1000	False	87.67	87.87	94.40	12.00	12.33	89.33	86.45	86.00	21	16	134	129
ViT + FW	Minmax	278	True	90.67	90.91	95.42	10.00	9.33	93.33	88.61	88.00	18	10	140	132
ViTFER	None	1000	False	87.00	87.21	93.44	12.67	13.00	88.67	85.81	85.33	22	17	133	128
ViTFER + FW	Standard	265	True	88.33	88.45	93.33	12.00	11.67	89.33	87.58	87.33	19	16	134	131
ViTSwin	None	1000	False	90.33	90.49	95.35	10.67	9.67	92.00	89.03	88.67	17	12	138	133
ViTSwin + FW	Minmax	163	True	92.67	92.81	95.29	8.67	7.33	94.67	91.03	90.67	14	8	142	136

Quick links

Search