Supervised Contrastive Learning for Generalizable and Explainable DeepFakes Detection

[Paper]
Ying Xu, Kiran Raja, Marius Pedersen

Introduction:

We propose a generalizable detection model that can detect novel and unknown/unseen DeepFakes using a supervised contrastive (SupCon) loss. We obtain the highest accuracy of 78.74% using proposed SupCon model and an accuracy of 83.99% with proposed fusion in a true open-set evaluation scenario where the test class is unknown at the training phase.

Framework:

How to use:

main_supcon.py
main_linear.py

Just remember two trains needed to be conducted.

Datalist

It is a .txt file that includes 'image_path label' every line. Here is an example:

FaceForensics++/original_sequences/youtube/c23/face_images/870/frame121.png 0
FaceForensics++/manipulated_sequences/Deepfakes/c23/face_images/979_875/frame1.png 1
...

Download model

Let me check if I have it.

Citing:

Please kindly cite the following paper, if you find this code helpful in your work.

@inproceedings{xu2022supervised,
  title={Supervised contrastive learning for generalizable and explainable deepfakes detection},
  author={Xu, Ying and Raja, Kiran and Pedersen, Marius},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={379--389},
  year={2022}
}

Contact:

Please feel free to contact ying.xu@ntnu.no, if you have any questions.