FDFtNet: Facing Off Fake Images using Fake Detection Fine-tuning Network.

Hyeonseong Jeon, Youngoh Bang, and Simon S. Woo

Overview of our framework.

Clone

git clone https://github.com/cutz-j/FDFtNet

(1) Setup

Install packages

• pip install -r requirements.txt

(2) Datasets

The dataset in the paper can be downloaded here.

• Deepfake: https://github.com/ondyari/FaceForensics
• Face2Face: https://github.com/ondyari/FaceForensics
• PGGAN dataset is not disclosed at the request of the dataset source.

Original dataset of the Deepfake, Face2Face can also be downloaded from FaceForensics

Preprocessing

# Crop the face parts to image from the videos by frame
cd dataset
# using MTCNN [1]
python preprocess_dataset.py -d youtube

python preprocess_dataset.py
 -d <dataset type, e.g., youtube, Deepfakes, Face2Face>

(3) Models

• SqueezeNet
• ResNetV2
• Xception
• ShallowNetV3 --> not disclosed at the request of the model source.

(4) Pretrain models

# Dataset needs to be downloaded.
# Choose networks above (squeezeNet, xception, resnetv2)

python pretrain.py --network xception --train_dir dataset --val_dir dataset

(5) Fine-tune

# Dataset needs to be downloaded.
# Choose networks above (squeezeNet, xception, resnetv2)
# Pre-trained model is needed.

python fdft.py --pt_model directory --network xception --ft_dir dataset --val_dir dataset

References

[1] Zhang, K., Zhang, Z., Li, Z., and Qiao, Y. (2016). Joint face detection and alignment using multitask cascaded convolutional networks. IEEE Signal Processing Letters, 23(10):1499–1503.