A Tensorflow Implementation of MesoNet: a Compact Facial Video Forgery Detection Network (https://arxiv.org/abs/1809.00888)
REQUIREMENTS :
Mesonet utilises Deepfake detection on Cropped Faces. For this a fast and accurate pytorch based tool is used to crop the faces.
git clone https://github.com/elliottzheng/face-detection
DATASET :
We use the Deepfake Dataset DFDC available publicly at Facebook. (https://ai.facebook.com/datasets/dfdc/)
The dataset has videos and its labels reside in JSON format for each file. For parsing this information and placing videos in respective folders , after downloading the dataset run the command below :
python filter_videos.py
TRAINING :
python train.py
STEPS :
