Skip to content

Latest commit

 

History

History
72 lines (60 loc) · 2.82 KB

README.md

File metadata and controls

72 lines (60 loc) · 2.82 KB
Raw

The Triple Robust Deepfake Detection Model

This is project code for SKT AI Fellowship 3 (2021)

Research Topic: Development of GAN-generated fake image detection technology

We aim to develop robust Deepfake detection model under the unseen environments as belows:

  • Deepfake method shift
  • Adversarial attack
  • Physical condition shift

Environment settings

conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch 
pip install matplotlib scikit-learn opencv-python albumentations tensorboard

Train model

python scripts/train_model.py <GPU index> <model name> <train dataset> <flag> --<other options>
ex) python scripts/train_model.py 0,1,2,3 EfficientNet-B4 FaceForensics,UDAFV,CelebDF mymodel --img_size 299 
ex) python scripts/train_model.py 0 EfficientNet-B4 CelebDF mymodel --phase_concat --oudefend --self_supervised

Test model

# single model
python scripts/train_model.py <GPU index> <model name> <seen dataset> <checkpoint path> --<other options>
python scripts/train_model.py 0 EfficientNet-B4 CelebDF ./checkpoints/mymodel/best_ckpt --phase_concat --oudefend --self_supervised

# ensenble model
ex) python scripts/test_model.py <GPU index> <None> <seen dataset> <None> --<other options> --ensemble 
--ensemble_ckpt_list  <checkpoint path 1> <checkpoint path 2> <checkpoint path 3> 
--ensemble_opt_list <option1_1:value/option1_2:value> <option2_1:value/option2_2:value/option2_3:value> <option3_1:value>

Available Models

  • Xception
  • Xception (SPSL)
  • Meso4, MesoInception4
  • EfficientNet
  • MobileNet-v2
  • MobileNet-v3
  • ResNet-50
  • WideResNet-50

Datasets

Reference

Spatial-Phase Shallow Learning: Rethinking Face Forgery Detection in Frequency Domain (CVPR 2021)
Adversarial Perturbations Fool Deepfake Detectors (IJCNN 2020)
Security of Facial Forensics Models Against Adversarial Attacks (ICIP 2020)
A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020)
DeepfakeUCL: Deepfake Detection via Unsupervised Contrastive Learning (arXiv 2021.04)
The Deepfake Detection Challenge Dataset (arXiv 2020.06)
FaceForensics : Learning to Detect Manipulated Facial Images (ICCV 2019)
Exposing Deep Fakes Using Inconsistent Head Poses (IEEE ICASSP 2019)
DeepFakes: a New Threat to Face Recognition? Assessment and Detection (arXiv 2018.12)
Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensics (CVPR 2020)