DeepFakes have been in the news for over a year now, mostly for all the wrong reasons. This repository is home to Pytorch implementations of various deepfake detection algorithms, tested on Google AI's Deep Fake Detection Dataset.
-
Data : Due to computational constraints, trained the MesoInception4 network on a subset of the DeepFake Detection dataset(363 videos from both real and manipulated classes).
Results : The data was trained for 30 epochs using all default hyperparamametes and network as mentioned in the original paper.
Training Accuracy: 93.75% Test Accuracy: 90.42%