Facebook, Microsoft, Amazon, and academics from several US universities have joined forces in an effort to fight against deepfakes, and have invested in the large Deepfake Detection Challenge which was launched today.
We wrote several months ago about the plans of Facebook to build such a challenge and the release of some preview datasets created specifically for it. Back then, Facebook announced that they will invest in this challenge to address the problem of deepfakes spread which has significant implications in determining the legitimacy of the information. According to them, the goal of such kind of challenge is to produce deepfakes detection technology that can be available to anyone. To achieve this goal Facebook AI put out a casting call to collect data with a crowdsourcing campaign in an ethical way.
Today, the challenge was launched as part of Google’s machine learning and data science platform Kaggle. The Deepfakes Detection Challenge will be open until March 31st, 2020 which is the final submission deadline and it offers more than 1 million dollars in prizes. The top Kaggle prize is 500 000$, followed by second and third prize with 300 000$ and 100 000$. Also, the fourth and the fifth entry will receive prizes of 60 000$ and 40 000$, respectively.
The participants will have access to the large deepfakes dataset collected by Facebook AI Research, which was split into a training set, public validation set, public test set, and a private test set. The challenge is already open and more details about the rules and requirements can be found in the challenge overview page.