Facebook, AWS, Microsoft and others launch coding challenge for deepfake detection
How can we detect deepfakes that have been created with deep learning methods such as GANs? Facebook, AWS and Microsoft joined forces to launch the Deepfake Detection Challenge (DFDC) that should encourage developers to research this issue. Winners can receive up to $500,000 USD.
Facebook announced a challenge to develop technologies for detecting deepfakes, titled the Deepfake Detection Challenge (DFDC). The announcement took place at NeurIPS 2019 (Conference on Neural Information Processing Systems), Dec 8-14, 2019.
Deepfakes are AI-generated videos that can falsely depict people saying or doing things that didn’t happen, and it’s not hard to imagine what harm this may cause. Beside the tech giants Facebook, Microsoft and Amazon Web Services (AWS), the challenge was also developed by The Partnership on AI Steering Committee on AI and Media Integrity and academics.
Let’s take a look at the requirements and prizes.
Deepfake Detection Challenge requirements
The competition is carried out via DFDC’s Kaggle page. It comes with four groups of datasets, namely a training set available outside Kaggle, a public validation set within Kaggle, a public test set that is completely withheld and a private test set that is privately held outside Kaggle.
The DFDC team offers some tips, e.g. ensuring the code can be re-run on any unseen test set, and warns not to create multiple accounts as these users will be excluded from participating.
Submissions must be made through Kaggle Notebooks, Kaggle’s cloud computational environment for machine learning code. Be sure to read the full instructions, as the workflow is claimed to be unique for this competition.
Deadlines and prizes
The entry deadline is March 3, 2020, which also marks the last day participants may join or merge teams. The final deadline for submissions is March 31, 2020. All deadlines are at 11:59:00 PM UTC on the respective day. The prizes for the five highest-ranked projects range from $500,000 USD to $40,000 USD.