Table 2. The performace of our method and other state-of-art methods on FaceForensics++ dataset.

Method	Raw		C23		C40
	ACC	AUC (%)	ACC	AUC (%)	ACC	AUC (%)
Xception (Chollet, 2017)	99.26	99.2	95.73	96.3	86.86	89.3
Face X-ray (Li et al., 2020a)	–	–	–	87.4	–	61.6
$F^3$Net (Qian et al., 2020)	99.95	99.8	97.52	98.1	90.43	93.3
Two-branch (Masi et al., 2020)	–	–	96.43	98.7	86.34	86.59
WDB (Jia et al., 2021)	99.74	99.78	96.95	99.6	88.96	92.97
FDFL (Li et al., 2021)	–	–	96.69	98.5	89.0	92.4
LRL (Chen et al., 2021)	99.87	99.92	97.59	99.46	91.47	95.21
M2TR (Wang et al., 2022)	99.50	99.92	97.93	99.51	92.89	95.31
RECCE (Cao et al., 2022)	–	–	97.06	99.32	91.03	95.02
GocNet (Guo et al., 2023c)	–	–	94.34	97.75	89.46	92.52
LDFnet (Guo et al., 2023b)	–	–	96.01	98.92	92.32	96.79
Our	99.62	99.87	97.98	99.64	92.92	94.35

Note:

Bold values refer to the best values.