Micro Expression Recognition via Dual-Stream Spatiotemporal Attention Network

Table 4

Comparison of DSTAN with other approaches (F1 : F1-Score).

Method	SMIC		CASME		CASME II
	Accuracy	F1	Accuracy	F1	Accuracy	F1
LBP-TOP [6]	0.44	0.41	0.37	0.33	0.47	0.35
LBP-SIP [7]	0.45	0.45	0.37	0.33	0.47	0.45
STLBP-IP [8]	0.58	—	—	—	0.56	—
STCLQP [9]	0.64	0.64	0.57	0.50	0.58	0.58
LCBP [10]	0.71	0.68	—	—	0.70	0.70
MDMO [11]	0.59	0.59	0.56	0.56	0.52	0.50
FDM [12]	0.55	0.54	0.56	0.49	0.46	0.41
Sparse MDMO [13]	0.71	0.70	0.75	0.75	0.67	0.69
Bi-WOOF [14]	0.62	0.62	—	—	0.59	0.61
CNN-LSTM [15]	—	—	—	—	0.61	—
3D-FCNN [16]	0.56	—	0.54	—	0.59	—
ELRCN-TE [17]	—	—	—	—	0.52	0.50
STRCN [18]	0.72	0.69	—	—	—	—
MERTA [21]	—	—	—	—	0.61	—
DSTAN	0.77	0.78	0.78	0.75	0.75	0.73