Google in collaboration with Jigsaw, have released a large dataset of visual deepfakes, which has been incorporated into the Technical University of Munich and University Federico II of Naples’ new FaceForensics benchmark, aimed at directly complimenting deepfake detection efforts.
Deepfakes have presented huge threats, with the latest AI capabilities, there are huge possibilities that are both exciting and horrifying at the same time. And the technology isn’t just limited to online platforms, just like any transformational technology, this has created some challenges.
While Google had earlier released a dataset of synthetic speech to develop high-performance fake audio detectors, in support of an international challenge. And the dataset has been downloaded by over 150 researchers and industry organizations as part of the challenge, which is also now freely available to the public.
These technologies would have been thought as impossible only a few years ago, now there are generative models capable of synthesizing hyper-realistic images, voice, music, or even video. And these models have been employed in a wide variety of applications, including text-to-speech applications, and generation of training data for medical imaging capabilities.
Google over the years have worked with both paid and consenting actors to provide hundreds of videos, while using publicly available deepfake generation methods, to recreated thousands of deepfakes from the videos.
The final results, comprising videos, both real and fake, will directly support deepfake detection efforts, and serve as part of the FaceForensics benchmark, free to the research community, and for use in developing synthetic video detection methods.
But the threat posed by the deepfake technology isn’t just limited to online platforms, if regulations and policies are not implemented to prevent the harm that technologies like DeepFakes are capable of doing, nothing else can stop it.