FaceForensics Benchmark
This table lists the benchmark results for the Binary Classification scenario.
Method | Info | Deepfakes | Face2Face | FaceSwap | NeuralTextures | Pristine | Total |
---|---|---|---|---|---|---|---|
DirechletEnsemble-Classifier | 0.955 | 0.956 | 0.932 | 0.967 | 0.992 | 0.973 | |
Top8-soft-withoutnt | 0.300 | 0.066 | 0.913 | 0.007 | 0.974 | 0.624 | |
Cancer | 0.964 | 0.781 | 0.942 | 0.780 | 0.952 | 0.903 | |
lcy2 | 0.900 | 0.073 | 0.845 | 0.053 | 0.950 | 0.679 | |
Beijing ZKJ | 1.000 | 0.920 | 0.961 | 0.880 | 0.948 | 0.941 | |
YjzV4 | 0.200 | 0.153 | 0.107 | 0.707 | 0.938 | 0.629 | |
ZAntiFakeBio | 1.000 | 0.920 | 0.971 | 0.907 | 0.936 | 0.940 | |
Leo | 1.000 | 0.861 | 0.971 | 0.853 | 0.922 | 0.917 | |
MixingExpert | 1.000 | 0.905 | 0.942 | 0.780 | 0.922 | 0.909 | |
Firefly | 0.955 | 0.730 | 0.874 | 0.667 | 0.920 | 0.855 | |
Yjz | 0.964 | 0.664 | 0.942 | 0.553 | 0.920 | 0.837 | |
FAKEDET | 0.964 | 0.832 | 0.922 | 0.687 | 0.918 | 0.877 | |
Yolo with Voting Mechanism | 0.855 | 0.606 | 0.709 | 0.587 | 0.916 | 0.796 | |
jiayoua | 0.327 | 0.803 | 0.117 | 0.740 | 0.908 | 0.723 | |
NoSenseAtAll | 0.982 | 0.905 | 0.951 | 0.827 | 0.908 | 0.908 | |
gg2 | 0.327 | 0.803 | 0.117 | 0.740 | 0.908 | 0.723 | |
RobustForensics | 0.991 | 0.891 | 0.951 | 0.807 | 0.904 | 0.902 | |
AllDataRes_EB5 | 0.964 | 0.818 | 0.845 | 0.647 | 0.900 | 0.852 | |
b7_e2_ensemble_chun | 0.927 | 0.759 | 0.806 | 0.687 | 0.900 | 0.842 | |
ENbsoftFW1111 | 0.982 | 0.891 | 0.922 | 0.787 | 0.898 | 0.892 | |
ATDETECTOR | 0.955 | 0.796 | 0.922 | 0.780 | 0.898 | 0.875 | |
winer_ensemble_correct2 | 0.873 | 0.745 | 0.816 | 0.560 | 0.896 | 0.814 | |
PredictFake | 0.973 | 0.847 | 0.913 | 0.820 | 0.894 | 0.887 | |
lcy | 0.964 | 0.883 | 0.893 | 0.573 | 0.890 | 0.850 | |
yangjiezhi-ensemble | 0.927 | 0.745 | 0.816 | 0.700 | 0.884 | 0.835 | |
Aquarius | 1.000 | 0.854 | 0.971 | 0.807 | 0.884 | 0.890 | |
StableForensics | 0.991 | 0.847 | 0.951 | 0.787 | 0.884 | 0.883 | |
effb7_newdataaug_v7_aligne_v2 | 0.945 | 0.810 | 0.942 | 0.613 | 0.884 | 0.846 | |
InTeLe_ | 1.000 | 0.796 | 0.922 | 0.720 | 0.884 | 0.864 | |
b7_514_e2_luo_yang_efficienet | 0.936 | 0.825 | 0.854 | 0.700 | 0.884 | 0.851 | |
en7_yjz | 0.973 | 0.898 | 0.922 | 0.720 | 0.880 | 0.873 | |
FullEfficientNet | 0.836 | 0.526 | 0.728 | 0.480 | 0.880 | 0.751 | |
ZFake | 0.945 | 0.737 | 0.864 | 0.693 | 0.880 | 0.838 | |
yjzenb_lowqulity | 0.973 | 0.905 | 0.932 | 0.727 | 0.880 | 0.876 | |
696_723_yjz | 0.964 | 0.818 | 0.893 | 0.740 | 0.878 | 0.860 | |
EnbFW | 0.973 | 0.912 | 0.951 | 0.727 | 0.878 | 0.878 | |
Balance | 0.918 | 0.861 | 0.883 | 0.733 | 0.876 | 0.858 | |
CONG | 0.955 | 0.810 | 0.854 | 0.780 | 0.874 | 0.858 | |
EfficienNetb7 | 0.973 | 0.905 | 0.922 | 0.720 | 0.868 | 0.868 | |
result_all_best_th0.85_517_ens | 0.945 | 0.847 | 0.854 | 0.740 | 0.866 | 0.852 | |
HKU_EfficientNet-B7_v2 | 0.845 | 0.606 | 0.806 | 0.447 | 0.864 | 0.758 | |
framelevel_all_compress_format | 0.782 | 0.124 | 0.155 | 0.173 | 0.860 | 0.575 | |
Yolo Object detection | 0.864 | 0.628 | 0.757 | 0.667 | 0.854 | 0.786 | |
shangci | 0.973 | 0.876 | 0.942 | 0.747 | 0.854 | 0.863 | |
KBNet | 0.909 | 0.774 | 0.854 | 0.793 | 0.852 | 0.839 | |
HRC | 0.827 | 0.613 | 0.573 | 0.660 | 0.848 | 0.757 | |
iouiwc-31 | 0.964 | 0.796 | 0.893 | 0.760 | 0.848 | 0.845 | |
EfficientB5_OHEM | 0.918 | 0.876 | 0.913 | 0.653 | 0.848 | 0.837 | |
LGSC_Lite | 0.955 | 0.839 | 0.874 | 0.773 | 0.844 | 0.848 | |
Yjz_jerry | 0.973 | 0.876 | 0.893 | 0.713 | 0.840 | 0.846 | |
focbe | 0.900 | 0.832 | 0.786 | 0.793 | 0.828 | 0.827 | |
IAMFAKE2 | 0.464 | 0.496 | 0.515 | 0.367 | 0.826 | 0.640 | |
IAMFAKE | 0.436 | 0.409 | 0.524 | 0.300 | 0.816 | 0.611 | |
unet+res | 0.882 | 0.832 | 0.796 | 0.767 | 0.812 | 0.814 | |
UltraVision | 0.918 | 0.847 | 0.932 | 0.733 | 0.810 | 0.828 | |
YJZ_419_513_ensemble | 0.964 | 0.832 | 0.951 | 0.780 | 0.808 | 0.839 | |
fakeface | 0.955 | 0.766 | 0.883 | 0.747 | 0.806 | 0.816 | |
effb7_newdataaug_v7_aligne | 0.973 | 0.825 | 0.961 | 0.660 | 0.804 | 0.820 | |
YJZ_ensemmble | 0.936 | 0.810 | 0.942 | 0.753 | 0.794 | 0.821 | |
EfficientB4_380 | 0.882 | 0.701 | 0.796 | 0.740 | 0.794 | 0.783 | |
IR-Capsule | 0.973 | 0.818 | 0.942 | 0.793 | 0.792 | 0.831 | |
RealFace | 0.945 | 0.766 | 0.864 | 0.813 | 0.790 | 0.815 | |
IR+unet | 0.909 | 0.847 | 0.845 | 0.787 | 0.790 | 0.816 | |
resnet50 | 0.527 | 0.504 | 0.408 | 0.387 | 0.786 | 0.620 | |
Xceptio InceptionRes Efficient | 0.964 | 0.825 | 0.864 | 0.827 | 0.768 | 0.816 | |
inceptionv1 | 0.655 | 0.350 | 0.544 | 0.333 | 0.768 | 0.610 | |
antifake | 0.936 | 0.803 | 0.942 | 0.687 | 0.764 | 0.795 | |
SCAN | 0.909 | 0.825 | 0.738 | 0.613 | 0.764 | 0.763 | |
yd dust face | 0.955 | 0.861 | 0.951 | 0.693 | 0.762 | 0.806 | |
framelevel | 0.873 | 0.737 | 0.835 | 0.760 | 0.760 | 0.777 | |
Inception Resnet V1 | 0.936 | 0.839 | 0.903 | 0.820 | 0.750 | 0.809 | |
Nika Dogonadze, Jana Obernosterer: Deep Face Forgery Detection. Advanced Deep Learning for Computer Vision Course at TUM | |||||||
SiameseDF | 0.864 | 0.723 | 0.709 | 0.667 | 0.750 | 0.742 | |
DFSpot-Ensemble | 0.900 | 0.766 | 0.825 | 0.700 | 0.750 | 0.769 | |
Siam-ensemble | 0.945 | 0.934 | 0.951 | 0.747 | 0.748 | 0.816 | |
simple policy | 0.955 | 0.869 | 0.942 | 0.387 | 0.744 | 0.751 | |
face single model | 0.964 | 0.869 | 0.942 | 0.460 | 0.738 | 0.760 | |
faceClassify1 | 0.945 | 0.854 | 0.981 | 0.793 | 0.730 | 0.806 | |
MesoNet | 0.873 | 0.562 | 0.612 | 0.407 | 0.726 | 0.660 | |
Darius Afchar, Vincent Nozick, Junichi Yamagishi, and Isao Echizen: Mesonet: a compact facial video forgery detection network. arXiv | |||||||
swap_classify | 0.909 | 0.759 | 0.864 | 0.747 | 0.724 | 0.767 | |
EfficientNet-b4 | 0.955 | 0.796 | 0.825 | 0.827 | 0.712 | 0.779 | |
EffB7_DataAug_v1 | 0.945 | 0.847 | 0.825 | 0.813 | 0.712 | 0.783 | |
EffB7_DataAug_v7 | 0.964 | 0.920 | 0.951 | 0.867 | 0.710 | 0.815 | |
AttentionExcep | 0.782 | 0.723 | 0.524 | 0.747 | 0.700 | 0.701 | |
LVLNet | 0.964 | 0.752 | 0.796 | 0.687 | 0.694 | 0.741 | |
xceptattention | 0.791 | 0.774 | 0.456 | 0.747 | 0.680 | 0.692 | |
EfficientNetTest | 0.709 | 0.445 | 0.631 | 0.387 | 0.674 | 0.599 | |
EffB8_FFDFDCDFCelebAISG | 0.382 | 0.409 | 0.447 | 0.433 | 0.660 | 0.539 | |
Xception_ForgeryNet | 0.818 | 0.496 | 0.544 | 0.407 | 0.632 | 0.591 | |
EfficientNetB7 | 0.836 | 0.847 | 0.874 | 0.747 | 0.630 | 0.725 | |
PPPP | 0.827 | 0.635 | 0.621 | 0.480 | 0.628 | 0.628 | |
Sentinel | 0.964 | 0.905 | 0.883 | 0.867 | 0.624 | 0.763 | |
EfficientNet-b7-Attention | 0.955 | 0.803 | 0.854 | 0.600 | 0.616 | 0.701 | |
eff-b7-att epc11 | 0.955 | 0.788 | 0.864 | 0.547 | 0.592 | 0.680 | |
MMIG-Net | 0.909 | 0.825 | 0.816 | 0.887 | 0.566 | 0.713 | |
GAEL-Net | 0.718 | 0.686 | 0.631 | 0.707 | 0.562 | 0.625 | |
Jae-Yong Baek, Yong-Sang Yoo, Seung-Hwan Bae: Generative Adversarial Ensemble Learning for Face Forensics. IEEE Access | |||||||
Forged face detection : fCNN | 0.791 | 0.642 | 0.709 | 0.513 | 0.544 | 0.597 | |
Aditi Kohli, Abhinav Gupta: Detecting DeepFake, FaceSwap and Face2Face facial forgeries using frequency CNN. Multimedia Tools and Application, Springer | |||||||
Xception | 0.964 | 0.869 | 0.903 | 0.807 | 0.524 | 0.710 | |
Andreas Rössler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, Matthias Nießner: FaceForensics++: Learning to Detect Manipulated Facial Images. ICCV 2019 | |||||||
XceptionNet Full Image | 0.745 | 0.759 | 0.709 | 0.733 | 0.510 | 0.624 | |
Andreas Rössler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, Matthias Nießner: FaceForensics++: Learning to Detect Manipulated Facial Images. ICCV 2019 | |||||||
Rahmouni | 0.855 | 0.642 | 0.563 | 0.607 | 0.500 | 0.581 | |
Nicolas Rahmouni, Vincent Nozick, Junichi Yamagishi, and Isao Echizen: Distinguishing computer graphics from natural images using convolution neural networks. IEEE Workshop on Information Forensics and Security, | |||||||
DFirt | 0.982 | 0.942 | 0.961 | 0.893 | 0.494 | 0.717 | |
YSNet | 0.564 | 0.584 | 0.592 | 0.427 | 0.492 | 0.513 | |
p-DARTS, generalized-cells | 0.791 | 0.730 | 0.816 | 0.720 | 0.478 | 0.618 | |
Jordi Moreno: Progressive Differentiable Architecture Search for DeepFake Detection. MSc Research Project, from Upf | |||||||
Bayar and Stamm | 0.845 | 0.737 | 0.825 | 0.707 | 0.462 | 0.616 | |
Belhassen Bayar and Matthew C. Stamm: A deep learning approach to universal image manipulation detection using a new convolutional layer. ACM Workshop on Information Hiding and Multimedia Security | |||||||
EfficientNet-v0 | 0.827 | 0.730 | 0.806 | 0.533 | 0.420 | 0.564 | |
Recasting | 0.855 | 0.679 | 0.738 | 0.780 | 0.344 | 0.552 | |
Davide Cozzolino, Giovanni Poggi, and Luisa Verdoliva: Recasting residual-based local descriptors as convolutional neural networks: an application to image forgery detection. ACM Workshop on Information Hiding and Multimedia Security | |||||||
Steganalysis Features | 0.736 | 0.737 | 0.689 | 0.633 | 0.340 | 0.518 | |
Jessica Fridrich and Jan Kodovsky: Rich Models for Steganalysis of Digital Images. IEEE Transactions on Information Forensics and Security | |||||||
eff-b7-v3 | 0.973 | 0.912 | 0.913 | 0.807 | 0.198 | 0.546 | |
Two-stream-SRM-RGB1 | 0.982 | 0.927 | 0.942 | 0.953 | 0.174 | 0.562 | |
SRTNet_formal_v2 | 0.982 | 0.956 | 0.951 | 0.960 | 0.066 | 0.514 | |