Table 3.

UAR and UF1 performance of different approaches under LOSO protocol on composite or individual datasets.

Approach	Composite dataset		CASME II		SMIC		SAMM
	UAR	UF1	UAR	UF1	UAR	UF1	UAR	UF1
LBP-TOP (Pfister et al., 2011)	0.5785	0.5882	0.5429	0.5026	0.5280	0.2000	0.4102	0.3954
LBP-SIP (Wang et al., 2014)	0.4681	0.4829	0.5281	0.5369	0.5142	0.4452	0.4169	0.4412
HOOF (Chaudhry et al., 2009)	0.5814	0.5982	0.5782	0.5874	0.5696	0.5574	0.5877	0.5639
MDMO (Liu et al., 2015)	0.5125	0.5635	0.5382	0.5492	0.4812	0.4926	0.5108	0.5021
ResNet18	0.6682	0.6715	0.6522	0.6428	0.6271	0.6542	0.6632	0.6743
CNN-LSTM (Wang et al., 2018)	0.3942	0.3852	0.4125	0.4113	0.4276	0.4150	0.3086	0.3020
DenseNet121	0.3414	0.4253	0.3334	0.4604	0.3518	0.2909	0.3374	0.5645
RCN-Best (Xia et al., 2020b)	0.7190	0.7466	0.6600	0.6584	0.8131	0.8653	0.6771	0.7647
TSCNN (Song et al., 2019)	0.5849	0.5923	0.6009	0.6124	0.5924	0.5839	0.6103	0.6083
MobileNetV2 (Sandler et al., 2018)	0.6425	0.6652	0.6328	0.6125	0.6368	0.6589	0.6236	0.6614
DeepViT (Zhou et al., 2021)	0.7025	0.7158	0.7001	0.6982	0.7152	0.7369	0.7114	0.6928
DeiT (Touvron et al., 2021)	0.6879	0.6731	0.6814	0.6994	0.6881	0.6970	0.7052	0.7028
MobileViT (Ours)	0.6981	0.7318	0.6997	0.7251	0.7356	0.7141	0.6781	0.7428

The bold values indicate the highest values under the particular metrics.