In an attempt to cope with the rising concerns over deepfake videos, Google has released a massive dataset of over 3000 deepfake videos, read an official blog post.
The search giant has done so in collaboration with Jigsaw. Their dataset is now a part of the FaceForensics benchmark developed by the Technical University of Munich and the University Federico II of Naples.
How Google pulled this off?
To create deepfake videos, Google contacted 28 paid and consenting actors and recorded hundreds of videos in various scenes. It then used publicly available deepfake algorithms to swap the actors faces.
Google says it will make further contributions to the dataset in the coming future as the deepfake technology improves over time. This is surely going to help interested researchers in detecting potential fake videos.
Why is this important?
The need for detecting deepfake videos has become more important as this technology has been transformed into a newly-found tool to manipulate videos for revenge porn, defaming people, and spreading fake news.
Earlier this month, Facebook, which has become a common platform for spreading deepfake videos, announced a Deepfake Detection Challenge with a $10 million prize money. The challenge involved a new dataset containing synthesized videos made with the help of paid actors.
Not just videos, AI is also being used to manipulate images as well. Recently we heard about the DeepNude AI app which can be used to virtually undress women in images. The app was taken down by its creator, but it was definitely an example of how far the deepfakes technology has come.