imadtoubal / multimodaldeepfakedetection

An open-source repository that provides the code for the research conducted at the NSF University of Missouri REU Summer 2020

MultimodalDeepfakeDetection

An open-source repository that provides the code for the research conducted at the NSF University of Missouri REU Summer 2020

Command to process landmarks

python landmark_preprocess.py path/to/processed/videos path/to/save/output

Input folder should be a folder containing top-level folders real and fake. These folders contain video-label-named folders which will contain faces in JPEG file format.

Command to process 1D DCT of Landmarks

python dct.py path/to/processed/video/ path/to/save/output

Input folder should be a folder containing top-level folders real and fake. These folders contain video-label-named folders in which landmark-named folders (mouth, nose, both-eyes) are hosted. This dct.py script will take these images and save them in numpy sequences of seq-size. See dct.py --help for more information.

Command to process LipNet Sequences of Mouth

python lipnet_sequence.py path/to/processed/video-landmarks/ path/to/save/output

Input folder should be a folder containing top-level folders real and fake.

input_folder
    - real
        - *.mp4
    - fake
        - *.mp4

These folders contain video-label-named folders in which landmark-named folders (mouth, nose, both-eyes) are hosted. This lipnet_sequence.py script will take the mouths in this directory and save them into torches and then send this sequence through the LipNet model and extract the final features before the fully connected layer. See lipnet_sequence.py --help for more information.