List your company
ResearchFormerly WizWand

Deepfake Detection on DFDC

100AUCFACTOR
74.93681.44387.9594.457Jan 18, 2022Oct 10, 2022Jul 2, 2023Mar 24, 2024Dec 14, 2024Sep 5, 2025May 29, 2026
Updated 5mo ago

Evaluation results

MethodLinks
FACTORScenario=In-Method, Train/Test=B/B2023.11100——
FACTORScenario=Zero-Day, Train/Test=A/B2023.11100——
FACTORScenario=In-Method, Train/Test=A/A2023.1199.9——
FACTORScenario=Zero-Day, Train/Test=B/A2023.1199.9——
FACTOREvaluation Protocol=Ref. Set, Reference Set Source=50% subset of training videos2023.1199.7——
F3NETScenario=In-Method, Train/Test=A/A2023.1197.4——
UCFScenario=In-Method, Train/Test=A/A2023.1197——
Aletheia2026.0496.6——
SPSLScenario=In-Method, Train/Test=A/A2023.1196.4——
FFDScenario=In-Method, Train/Test=A/A2023.1196——
DF40 baseline2026.0496——
XceptionScenario=In-Method, Train/Test=A/A2023.1195.7——
LAA-Net2026.0495.5——
RECCEScenario=In-Method, Train/Test=A/A2023.1195——
UCFScenario=In-Method, Train/Test=B/B2023.1194.9——
Xception baseline2026.0494.1——
XceptionScenario=In-Method, Train/Test=B/B2023.1193.2——
F3NETScenario=In-Method, Train/Test=B/B2023.1189.4——
RECCEScenario=In-Method, Train/Test=B/B2023.1188.6——
F3NETScenario=Zero-Day, Train/Test=B/A2023.1187.9——
M2F2-Det (w/ SBI)Training set (Real)=true, Training set (Fake)=false2025.0387.8——
FreqBlenderVenue=NeurIPS24, Training set (Real)=true, Training set (Fake)=false2025.0387.56——
ForAdaTraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0487.2——
ForAdaInput=Frame, Backbone=CLIP ViT-L/14, Training Dataset=FF++2025.0887.2——
GenD (CLIP)Input=Frame, Backbone=CLIP ViT-L/14, Training Dataset=FF++2025.0887.1——
LAA-Net (w/ SBI)Venue=CVPR24, Training set (Real)=true, Training set (Fake)=false2025.0386.94——
LAA-NetInput=Frame, Backbone=EFNB4, Training Dataset=FF++2025.0886.9——
HAADTraining Dataset=FF++, Evaluation Protocol=Cross-dataset Evaluation2026.0586.8——
SePLTraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0486.6——
SeeableVenue=ICCV23, Training set (Real)=true, Training set (Fake)=false2025.0386.3——
ViT-Forensics2025.1186.2——
AUNetVenue=CVPR23, Training set (Real)=true, Training set (Fake)=false2025.0386.16——
SBIVenue=CVPR22, Training set (Real)=true, Training set (Fake)=false2025.0386.15——
FFDScenario=In-Method, Train/Test=B/B2023.1185.7——
INSIGHT2025.1185.7——
ForAdaTraining Data=FF++, Evaluation Protocol=Video-level2026.0585.6——
FS-VFMTraining Data=FF++, Evaluation Protocol=Video-level2026.0585.5——
CNN Forensics2025.1184.9——
EffortTraining Data=FF++, Evaluation Protocol=Video-level2026.0584.8——
GenD (DINO)Input=Frame, Backbone=DINOv3 ViT-L/16, Training Dataset=FF++2025.0884.7——
SPSLScenario=Zero-Day, Train/Test=A/B2023.1184.4——
OursTraining Dataset=FF++ c23, Evaluation Protocol=Protocol-12024.0884.3——
EffortTraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0484.3——
P&PInput=Video, Backbone=CLIP ViT-L/14, Training Dataset=FF++2025.0884.3——
EffortInput=Frame, Backbone=CLIP ViT-L/14, Training Dataset=FF++2025.0884.3——
EffortDetector Type=Binary detector2026.0584.2——
FakeRadarBackbone=Vision Transformer, Input Type=Video2025.1284.1——
X^2-DFDTraining Dataset=FF++, Evaluation Protocol=Cross-dataset Evaluation, Re-implemented=true2026.0584.1——
HFRTraining Dataset=FF++ [7]2026.0384.07——
EffortTraining Dataset=FF++, Evaluation Protocol=Cross-dataset Evaluation, Re-implemented=true2026.0584——
EffNetB4Scenario=In-Method, Train/Test=A/A2023.1183.5——
CDFATraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0483——
OursTraining Dataset=FaceForensics++ [48], Evaluation Protocol=cross-dataset2026.0582.75——
OursDetector Type=Language detector2026.0582.75——
DFD-FCGInput=Video, Backbone=CLIP ViT-L/14, Training Dataset=FF++2025.0881.8——
SPSLScenario=In-Method, Train/Test=B/B2023.1181.7——
GenD (PE)Input=Frame, Backbone=PEcoreL, Training Dataset=FF++2025.0881.6——
CLIP Global2025.1181.4——
UCFScenario=Zero-Day, Train/Test=A/B2023.1181.3——
UDDTraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0481.2——
UDDInput=Frame, Backbone=CLIP ViT-B/16, Training Dataset=FF++2025.0881.2——
GenD-PETraining Data=FF++, Evaluation Protocol=Video-level2026.0581.1——
BlenDTraining Data=25k ScaleDF (real) + SBI fakes, Evaluation Protocol=Video-level2026.0581——
RepDFDTraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0480.9——
VLFFDTraining Dataset=FF++, Evaluation Protocol=Cross-dataset Evaluation, Re-implemented=true2026.0580.9——
FSFMTraining Data=FF++, Evaluation Protocol=Video-level2026.0580.9——
DSMTraining Dataset=FF++ [7], Publication Year=20252026.0380.81——
UCFVenue=ICCV23, Training set (Real)=true, Training set (Fake)=true2025.0380.5——
UCFBackbone=Xception, Input Type=Frame2025.1280.5——
LTTDBackbone=Vision Transformer, Input Type=Video2025.1280.4——
RAEInput=Frame, Backbone=ViT-B/16, Training Dataset=FF++2025.0880.2——
F3NETScenario=Zero-Day, Train/Test=A/B2023.1179.8——
HybridTraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0479.6——
UCLVLMTraining Dataset=FaceForensics++ [48], Evaluation Protocol=cross-dataset2026.0579.12——
UCLVLMDetector Type=Language detector2026.0579.12——
SLADDTraining Data=F2F2022.0378.7——
GradCAM2025.1178.5——
TALL++Input=Video, Backbone=Swin-B, Training Dataset=FF++2025.0878.5——
FFDScenario=Zero-Day, Train/Test=A/B2023.1177.9——
EFFTTraining Dataset=FF++ [7], Publication Year=20252026.0377.65——
GM-DFTraining Dataset=FF++ [7], Publication Year=20242026.0377.23——
SLADDTraining Data=DF2022.0377.2——
SLADDVenues=CVPR 2022, Training Dataset=FF++ c23, Evaluation Protocol=Protocol-12024.0877.2——
SLADDTraining Dataset=FaceForensics++ (c23), Evaluation Protocol=Cross-dataset2026.0477.2——
LSDABackbone=EfficientNet, Input Type=Frame2025.1277——
LSDAInput=Frame, Backbone=EFNB4, Training Dataset=FF++2025.0877——
ProDetInput=Frame, Backbone=EFNB4, Training Dataset=FF++2025.0877——
TALLBackbone=Swin Transformer, Input Type=Video2025.1276.8——
DAIDTraining Dataset=FF++, Evaluation Protocol=Cross-dataset Evaluation, Re-implemented=true2026.0576.8——
Two-branchTraining Dataset=FaceForensics++, Evaluation Protocol=Cross-dataset generalization2022.0176.7——
NACOBackbone=Vision Transformer, Input Type=Video2025.1276.7——
NACOInput=Video, Backbone=ViT-B/16, Training Dataset=FF++2025.0876.7——
LocalRLVenue=AAAI21, Training set (Real)=true, Training set (Fake)=false2025.0376.53——
RealForensicsTraining Dataset=FaceForensics++, Evaluation Protocol=Cross-dataset generalization2022.0175.9——
UCFTraining Dataset=FF++(C23), Evaluation Protocol=Cross-Dataset2023.1175.9——
SeeABLEVenues=ICCV 2023, Training Dataset=FF++ c23, Evaluation Protocol=Protocol-12024.0875.9——
SeeABLEBackbone=EfficientNet, Input Type=Frame2025.1275.9——
RealForensicsBackbone=3D ResNet, Input Type=Video2025.1275.9——
RealForensicsInput=Video, Backbone=Modified CSN, Training Dataset=FF++2025.0875.9——
SeeABLEDetector Type=Binary detector2026.0575.9——