Open material developed for the project DeepFakeChain, funded under the H2020 TruBlo project.
The recent advancements in artificial intelligence (AI) have brought us deepfakes, i.e., highly photorealistic fake multimedia content that is difficult or impossible for the human eye to distinguish. While deepfakes offer attractive opportunities, e.g., for the entertainment sector, they raise concerns over their potential usage for misinformation, propaganda, fraud, unethical pornography and other harmful intents. To mitigate these intents, the algorithmic detection of deepfakes has been proposed, two obstacles exist however. On one hand, the current algorithms based on AI are not sufficiently reliable. On the other hand, we still need the contextual understanding of humans to verify such content.
The DeepFakeChain project addresses both points. First, it seeks to develop new algorithms for the automatic detection of deepfake content based on ensemble techniques, which forms the purely scientific part of the project. Second, it seeks to create a platform for the collaborative verification of multimedia content from human users and algorithms. This platform targets primarily journalists and fact-checkers but can be useful to other users like content moderators.
The DeepFakeChain project ran for 9 months, from August 2022 to April 2023. It has produced 4 deliverables, the 4th of which is a scientific publication describing the developed algorithms for deepfake detection. All deliverables and the code supporting the aforementioned publication can be found in this repository.
deliverables contains the deliverables produced during the project's duration.
code contains the code behind the project's publication.
short-overview.pdf is a brief presentation of the DeepFakeChain platform, as presented to TruBlo.
platform-demo.mp4 is a demo video of the developed platform.