linyuzhen / source-id-tracker

Source-ID-Tracker

Source ID Tracker: Source Face Identity Protection in Face Swapping in IEEE International Conference on Multimedia and Expo (ICME), 2022.

Pipeline

Preprocessing

Extracting face images with MTCNN:

python ext_ff.py

Generating face landmarks with dlib:

python save_dlib_mask/generate_landmarks_dlib_ff_df_save_img.py

Splitting the dataset and getting the txt:

python get_txt.py

Training

Training the Source-ID-Tracker model:

python train.py

Inference

1. Revealing source images from fake images

python generate_test_images.py
python calculate_id_sim.py
python calculate_metric.py
python calculate_id_sim_perturbation.py

2. Revealing source images from fake images of simulated distortions

python generate_test_images.py
python generate_perturbation.py
python Revealed_perturbation.py 
python calculate_id_sim_perturbation.py

Reference

@inproceedings{lin2022source,
  title = {Source-{{ID-Tracker}}: {{Source Face Identity Protection}} in {{Face Swapping}}},
  shorttitle = {Source-{{ID-Tracker}}},
  booktitle = {2022 {{IEEE International Conference}} on {{Multimedia}} and {{Expo}} ({{ICME}})},
  author = {Lin, Yuzhen and Chen, Han and Maiorana, Emanuele and Campisi, Patrizio and Li, Bin},
  year = {2022},
  pages = {01--06},
  issn = {1945-788X}
}