FakeTalkerDetect

Description

Deepfake Talking Head Detection with Siamese Networks for Unbalanced learning

paper: http://openaccess.thecvf.com/content_ICCVW_2019/papers/HBU/Jeon_FakeTalkerDetect_Effective_and_Practical_Realistic_Neural_Talking_Head_Detection_with_ICCVW_2019_paper.pdf

Dataset

Dataset: https://skku0-my.sharepoint.com/:f:/g/personal/byo7000_skku_edu/EjvIqW2O5UdGizE87UEKdFIBrGBC4qksJtUKLTscnjnhBQ?e=xTV2Tx

The talking-heads model was trained using the VoxCeleb2 dataset. The talking-headas model generates images. Talking-Heads GAN model from: https://github.com/grey-eye/talking-heads

RUN

run main.py

Extra

Test_ratio_result.ipynb

• real-sample proportion in test set experiment

Extra.Triplet_loss.ipynb

• Triplet loss experiment. * it is experimented in 50:50 proportion test set.