FD²Foremer: Thinking Face Forgery Detection in Midfrequency Geometry Details

Table 5

Results (%) of distillation for a sequence of selected . All trained with the image and the displacement map as inputs on low quality FF++. The metric is ACC.

Teacher	Student		Testing data (ACC)
			FF++(LQ)	DFDC
DF^2Former(swin)	DF^2Former(meta)	0	81.85	85.43
DF^2Former(swin)	DF^2Former(meta)	64	82.41	85.71
DF^2Former(swin)	DF^2Former(meta)	128	82.27	86.54
DF^2Former(swin)	DF^2Former(meta)	192	82.68	86.37
DF^2Former(swin)	DF^2Former(meta)	256	82.73	86.72

The metric is ACC. Best results are shown in bold.