Channel-Wise Spatiotemporal Aggregation Technology for Face Video Forensics

Table 5

Binary classification accuracy (%) (higher is better) of different backbones on frames.

Model	DF	F2F	FS	NT	DFDC-P	Celeb-DF
EfficientNet B0 [5]	99.31	99.69	99.53	99.13	81.97	93.97
Xception [22]	99.22	99.62	99.56	99.00	80.75	94.84
Inception V3 [23]	98.84	99.78	99.47	98.24	79.72	66.19
MobileNet V1 [24]	99.16	98.75	99.53	98.47	79.09	66.69
EfficientNet B0(w/o skip)	83.56	58.62	58.84	60.94	76.31	66.66
Xception (w/o skip)	94.91	58.80	64.62	53.91	65.44	67.50