|
Hsu, Zhuang & Lee (2020)
|
CNN concatenated to CFFN |
Image |
CelebA, DCGAN WGAN WGAN-GP, least squares GAN PGGAN. |
|
Chintha et al. (2020)
|
Convolutional bidirectional recurrent LSTM network |
Videos |
FaceForensics++ and Celeb-DF (5,639 deepfake videos) and the ASVSpoof Access audio dataset. |
|
Agarwal et al. (2020)
|
CNN |
Videos |
Four in-the-wild lip-sync deep fakes from Instagram and YouTube (www.instagram.com/bill posters ukand youtu.be/VWMEDacz3L4). |
|
Fernandes et al. (2020)
|
ResNet50model [102], pretrained on VGGFace2 |
Videos |
VidTIMIT and two other original datasets obtained from the COHFACE and Deepfake TIMIT datasets. |
|
Sabir et al. (2019)
|
Spatiotemporal features with RCN |
Videos |
FaceForensics++ dataset, including 1,000 videos. |
|
Xuan et al. (2019)
|
DCGAN, WGAN-GP and PGGAN. |
Images |
CelebA-HQ, DCGAN, GAN-GP and PGGAN |
|
Yang, Li & Lyu (2019)
|
SVM |
Videos/Images |
UADFV consists of 49 deepfake videos, and 252 deepfake images from DARPA MediFor GAN Image/Video Challenge. |
|
Nguyen, Yamagishi & Echizen (2019)
|
Capsule networks |
Videos/Images |
The Idiap Research Institute replayattack, facial reenactment FaceForensics. |
|
Afchar et al. (2018)
|
CNN |
Videos |
Deepfake one constituted from onlinevideos and the FaceForensics one created by the Face2Face approach. |
|
Güera & Delp (2018)
|
CNN and LSTM |
Videos |
A collection of 600 videos obtained from multiple websites. |
|
Li, Chang & Lyu (2018), Li & Lyu (2019)
|
LRCN |
Videos |
Consists of 49 interview and presentation videos, and their corresponding generated deepfakes. |
