FaceForensics is a video dataset containing more than 500,000 images of human faces from 1,004 videos. It can be used to study image or video forgery. All videos are downloaded from Youtube and edited into continuous short video clips, most of which involve frontal faces.

There are two versions of this dataset:

• Source-to-Target: The authors reenacted over 1,000 videos using new facial expressions extracted from other videos. This can be used to train a classifier to detect fake images or videos.

• Selfreenactment: The authors use Face2Face to input the same facial expression and reenact it in the video to obtain paired videos. This can be used to train supervised generative refinement models.