Table 2

Comparison of detection performance of several Deepfake detectors on the second genreation datasets under cross-domain training and with AUC as the performance metric. The AUC results of DefakeHop anad DefakeHop++ in both frame-level and video-level are given. The best and the second-best results are shown in boldface and underbared, respectively. Furthermore, we include results of DefakeHop and DefakeHop++ under the same-domain training in the last 4 rows. The AUC results of benchmarking methods are taken from [21] and the number of parameters are from https://keras.io/api/applications. Also, we use a to denote deep learning methods and b to denote non-deep-learning methods.

2nd Generation
MethodModelCeleb-DF v1Celeb-DF v2#param
Two-stream [44]InceptionV3a55.7%53.8%23.9M
Meso4 [1]Designed CNNa53.6%54.8%28.0K
MesoInception4 [1]Designed CNNa49.6%53.6%28.6K
HeadPose [37]SVMb54.8%54.6%
FWA [20]ResNet-50a53.8%56.9%25.6M
VA-MLP [23]Designed CNNa48.8%55.0%
VA-LogReg [23]Logistic Regressionb46.9%55.1%
Xception-raw [27]XceptionNeta38.7%48.2%22.9M
Xception-c23 [27]XceptionNeta65.3%22.9M
Xception-c40 [27]XceptionNeta65.5%22.9M
Multi-task [25]Designed CNNa36.5%54.3%
Capsule [26]CapsuleNeta57.5%3.9M
DSP-FWA [19]SPPNeta64.6%
Multi-attentional [43]Efficient-B4a67.4%19.5M
Ours (Frame Level)DefakeHop++b56.30%60.5%238K
Ours (Video Level)DefakeHop++b58.15%62.4%238K
Ours (Trained on Celeb-DF, Frame Level)DefakeHopb93.1%87.7%42.8K
Ours (Trained on Celeb-DF, Video Level)DefakeHopb95.0%90.6%42.8K
Ours (Trained on Celeb-DF, Frame Level)DefakeHop++b95.4%94.3%238K
Ours (Trained on Celeb-DF, Video Level)DefakeHop++b97.5%96.7%238K

or Create an Account

Close Modal
Close Modal