At NeurIPS 2019 this week Facebook unveiled its Deepfake Detection Challenge (DFDC), along with a new deepfake-specific dataset containing more than 100,000 videos.
Facebook AI Vice President Jerome Pesenti said he is “particularly excited about this project because it combines two elements that have been so effective in catalyzing AI research in other areas: an open challenge so researchers everywhere can compete and compare their results, and a large-scale, high-quality data set built expressly for this use case. Deepfakes are a rapidly evolving challenge, similar to spam, phishing, and other adversarial threats, and rapid progress will require contributions from experts across the AI community.”
Joining Facebook in the fight against manipulated media are Amazon Web Services (AWS), Microsoft, the Partnership on AI, Microsoft and representatives from leading academic institutions and media organizations.
“Deepfake” is a portmanteau of deep learning and fake. The seemingly magic media synthesis technology can swap faces from one person to another with high realism using machine learning techniques such as generative adversarial networks (GANs). Deepfakes caught the media spotlight in 2017 when a Redditor used GAN to create a series of fake celebrity porn videos that went viral. Actress Scarlet Johansson, a deepfake target, lashed out at the trend in a Washington Post interview: “the internet is a vast wormhole of darkness that eats itself.”
Early this year Facebook was criticized for failing to remove a viral video that had been manipulated to make US House Speaker Nancy Pelosi sound drunk. And then in June, the AI-powered app DeepNude went on sale, enabling users to virtually disrobe images of women. The app ignited a protest storm and was shut down by the developers.
The DeepNude fallout saw researchers such as machine learning pioneer Andrew Ng denouncing the misuse of AI technologies, and a commitment from Facebook to combat deepfakes, realized now in the DFDC.
DFDC Project leader Cristian Canton Ferrer says large scale “close-to-reality” datasets can play an important role in detecting deepfakes. Facebook created the new 100,000 video DFDC dataset from scratch, using paid actors in realistic scenarios. The highly diverse actors are 54 percent female and 46 percent male and in a variety of poses. The videos have diverse settings and background elements, and annotations describing whether they were manipulated by AI.
Facebook also announced that the online data science and machine learning community Kaggle, a subsidiary of Google, will host the DFDC challenge and leaderboard.
Facebook has dedicated US$10 million in awards and grants for the DFDC, which runs through March 31. First Prize is US$500,000.
See the DFDC webpage for more information or to enter the challenge.
Author: Yuqing Li | Editor: Michael Sarazen
It’s curious. Now it’s possible to make deepnude photos in seconds
http://nudify.cc/