Deepfake Detection

This repo will not be updated. We will public our new version at https://github.com/TARTRL/Deepfake_Dection

Description

Use deep models to detect fake images.

We trained an EfficientNet model to detect deepfake images. Check out this repo to know more about deepfake images.

• Real image:

• Fake image:

Installation

pip install -r requirements.txt

• Pytroch

Usage

• test images using trained model

You need download the pretrained model from BaiduYun:

download link: https://pan.baidu.com/s/16NIV5BVUITKwolQzPbj9Zw password:r3i8

And put the model(model_half.pth.tar) under ./models. Then you can run python script as below:

python test_images.py ./test_images/0.png ./test_images/1.jpeg

• train on your own dataset

The training code is to be released.

Cite

Please cite our code if you use this code or our models in your own work:

@misc{deepfake_detection,
  title={Deepfake Detection},
  author={Huang, Shiyu and others},
  howpublished={\url{https://github.com/huangshiyu13/deepfake_detection}},
  year={2020}
}

Author

Shiyu Huang ([email protected])

If you want to train or test your dataset using this code, please feel free to contact me.

Acknowledgement

This code is based on https://github.com/rwightman/pytorch-image-models