Face Forgery Detection with Long-Range Noise Features and Multilevel Frequency-Aware Clues

Table 6

Ablation studies of MFAM were carried out in the CNN-based branch within intra-dataset (FF++(40)) and cross-dataset (Celeb-DF) scenarios.

ID	Variant			FF++(c40)		Celeb-DF
	EfficientNetB4	Data preprocessing	MFRB	Acc	AUC	AUC
1				85.14	87.12	64.12
2		HF		86.52	88.13	65.23
3		HF + MF		86.65	88.31	65.81
4		HF + MF + LF		85.71	87.16	64.22
5		HF + MF		87.22	88.41	66.32

The best performances are marked with bold fonts.