Google releases new database to aid in deepfake detection
Many people are concerned about new advents in technology like the ability to create deepfakes that look so realistic it can be difficult to tell if the videos and images are the people they appear to be at a glance. Google wants to make it easier to know if the videos are real or deepfakes and is contributing to deepfake detection research. Google says that while many deepfakes are meant to be humorous, there is the chance that they could be harmful to individuals and society.Google says that it considers these issues to be severe and notes that it has in the past released a dataset of synthetic speech in support of an international challenge to develop fake audio detectors. Google says that dataset was downloaded by more than 150 research and industry organizations as part of the challenge. That same dataset is now available to the public.
Google has this week launched a new dataset of visual deepfakes that is in collaboration with Jigsaw. The new dataset has been incorporated into the Technical University of Munich and University Federico II of Naples' FaceForensics benchmark. The integration of the dataset into the benchmark will help to identify fakes and can be downloaded on the FaceForensics GitHub page.
Google says that the field is moving quickly and it will add to the dataset as deepfake technology evolves. Google also promises to continue to work with partners in the space. Google notes that it "firmly" believes in supporting a thriving research community and mitigating potential harms from misuse of synthetic media.
The new release of the FaceForensics dataset is called an "important step" in that direction. The deepfakes that were created for the dataset used actors, and while some of the fakes look very much like the actor, some of them look very different.