Pictures can be worth more than a thousand words in world of AI
Video image recognition the next artificial intelligence frontier
Posting a video that goes viral may get you your 15 minutes of fame. But David McNab has discovered there’s a way that moment can pay dividends for an amateur videographer.
It all boils down to video management rights. It’s something a vast majority of folks posting on social media don’t even think about, and for those who do, it’s usually much too complicated to master.
When McNab posted a video on his YouTube channel of refugee children tobogganing, it went viral. He was contacted by Rumble, a Toronto-based video rights platform that helps content creators monetize and distribute their video content. “They talked about copyright, licensing and permissions,” McNab says. “Because some people had started taking videos from my channel without permission, I got interested.”
Since posting his content on Rumble, a number of McNab’s videos have made their way to a global audience, including advertisers and media companies, who now pay for the rights to use them. Over the past 18 months, he says he has earned over $18,500 US, and monthly revenue rarely dips below $1,000. “If those videos were on YouTube, I would have only gotten about $20.”
A major issue for all types of creators is getting some value for their video content, says Chris Pavloski, founder and CEO at Rumble Inc. in Toronto. “They are either underpaid or people are stealing content they see posted on different platforms like Facebook or YouTube. We set out to solve the problem of how to manage and represent content. To do that, we started building processes and intelligence into licensing technology automatically.”
Established in 2013, the Rumble platform now draws more than 250 million views per month, has more than 100 million active users and 60,000 content creators, and boasts more than 300,000 social video assets.
Machine learning and artificial intelligence play a big part in being able to do this type of mass licensing on a large scale. The critical capability that has made it happen is the recent advancements of AI in vision recognition, Pavloski says.
“To get more intelligence you need visualization of video. Social video is the No. 1 way to achieve that because there is so much out there. With it you can create millions of models for machine learning to interpret for licensing, verification and protection of content, which allows us to distribute and monetize that content. Think of us as YouTube, except with an AI layer that licenses and distributes content intelligently.”
The importance of vision recognition should not be underestimated, Pavloski says. “Things are going to change pretty rapidly. Video is seen by many as the next major hurdle for AI.”
One person you don’t have to convince is Matthew Zeiler, cofounder and CEO of Clarifai in New York. A machine-learning specialist who studied under Geoffrey Hinton and other notable AI gurus, the Manitoba native has dedicated his research and development to building an image- and video-recognition platform for developers.
He launched his first image-recognition API (application program interface) in 2013. Right out of the gate, they entered an ImageNet image classification competition, winning the top five places. “It was a great way to kick off image recognition,” he says. “We got a lot of interest from press, investors and customers.” In 2015, Clarifai became the first to launch a video API.
As the technology evolves, the platform recognizes an ever-growing catalogue of visual images, from animals and plants to emotions, he explains. “It can take pixels and predict things automatically, which is something that is useful across all different industries such as food, wedding, and travel.”
Trivago, for example, uses the Clarifai platform to organize over 10 million listings by visual cues such as an ocean view or gym facilities. “If you want to search the best pool in Jamaica, it will compare images for you rather than making you click on every result.”
Image recognition can also be customized to focus on specific elements such as celebrities, age, gender and multiculturalism, Zeiler notes. The next frontier for Clarifai is running neural networks on cellphones.