Facebook AI Lumos Can Find Your Photos Even Those You Are Not Tagged In
Since Facebook is in the business of photos, it is not surprising to find it constantly developing products involving image recognition. This was recently demonstrated after the company announced that its artificial intelligence technology Lumos can now search for pictures not just based on dates, places, and tags. Rather, the technology can also find specific photos just because it understands what's in them.
Better Image Recognition Technology
If you type your name, for example, to trigger a Facebook photo search, results will start pulling images that you are not tagged in but contain your face. The same is true for searched things such as food and even actions.
Facebook is now able to do this by weaning itself away from text-driven technology. This means that the company is no longer relying solely on tags, labels, and descriptions that users enter into the system. The downside of this old method is that those photos that are not sufficiently tagged and captioned will not surface when searched.
"That's changing because we've pushed computer vision to the next stage with the goal of understanding images at the pixel level," Facebook said in an official statement. "This helps our systems do things like recognize what's in an image, what type of scene it is, if it's a well-known landmark, and so on."
Facebook's Lumos AI
At the heart of this initiative is Facebook's proprietary AI called Lumos. The company's engineers are developing and training it so it can better answer questions involving the elements and objects found in an image.
The AI technology draws from its deep-learning engine so it can attach more meaning to each photo in the Facebook system. The analogy is that Facebook's new technology can "see" the images as opposed to merely crawl their descriptions so that users can quickly find images even if they forgot their details.
Lumos's power can be demonstrated in the way it can process billions of photos to predict a set of concepts and recognize semantic meaning. As the tool rolls out, users can expect its impact on the way search image results are ranked.
According to Facebook, Lumos has been built on top of another technology called FBLearner Flow. This is a general service platform where Facebook's other AI initiatives are tested and trained. Interestingly, the latest AI image search tool emerged out of Facebook's desire to articulate photos and videos to the visually impaired.
Lumos is still getting developed with the company stressing that its engineers are merely scratching the surface. This should mean that the AI will still get better and more powerful over time, which should be aligned with the current trend at Facebook at the moment. If one recalls, the company has also turned to AI to solve the proliferation of fake news on its network. Even the Chan Zuckerberg Initiative has purchased its own AI startup to help the organization sort millions of scientific papers.