[Paper]
Ying Xu, Kiran Raja, Marius Pedersen
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.
main_supcon.py
main_linear.py
Just remember two trains needed to be conducted.
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
...
Let me check if I have it.
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}
}
Please feel free to contact ying.xu@ntnu.no, if you have any questions.