Facebook uses first-person videos to train future AIs
One of the obvious goals of almost every computer vision project is to enable a machine to see and perceive the world as a human does. Today Facebook started talking about Ego4D, its own effort in the field, for which it created a huge new dataset to train future models. In a statement, the company said it had recruited 13 universities in nine countries that had collected 2,200 hours of footage from 700 participants. This footage was captured from the user's perspective that can be used to train these future AI models. Kristen Grauman, lead researcher at Facebook, says this is the largest collection of data created specifically for this focus.
sunglasses, sunglasses: Ray-Ban Stories. © Karissa Bell Ray-Ban stories.
The footage focused on a range of common experiences in human life, including social interaction, hand and object manipulation, and predicting what will happen. For the social network, this is a big step towards better computing experiences, which until now have always been about getting data from the perspective of the viewer. Facebook has announced that the datasets will be released in November "for researchers who sign Ego4D's data usage agreement". And for the next year, researchers outside of this community will be challenged to better train machines to understand what exactly people are doing in their lives.
Of course there is the aspect that Facebook, which now has a camera glasses partnership with Ray Ban, would like to improve its own skills in the future. You probably already know the dangers of this potential surveillance and why someone might be a little suspicious of the announcement.