Multisemantic Path Neural Network for Deepfake Detection

Table 1

Accuracy comparison of multiscale detail enhancement methods for datasets with different compression rates before and after introduction.

	Acc (%) on FF++(HQ)				Acc (%) on FF++(LQ)
	DeepFakes	Face2Face	FaceSwap	NeuralTextures	DeepFakes	Face2Face	FaceSwap	NeuralTextures
Without detail enhancement	99.73	99.09	99.17	91.2	93.83	91.15	92.47	74.41
With detail enhancement	99.54	98.92	98.86	91.3	94.25	91.25	91.74	74.52

The bold values indicate the better results in the two experiments.