FakeBuster: A DeepFakes Detection Tool for Video Conferencing Scenarios.

摘要: This paper proposes a new DeepFake detector FakeBuster for detecting impostors during video conferencing and manipulated faces on social media. is standalone deep learning based solution, which enables user to detect if another person's or spoofed meeting. tool independent of solutions has been tested with Zoom Skype applications. It uses 3D convolutional neural network predicting segment-wise fakeness scores. The trained combination datasets such as Deeperforensics, DFDC, VoxCeleb, deepfake videos created using locally captured (for scenarios) images. leads different environments perturbations in the dataset, improves generalization network.