All about Meero's AI with Juliette Chataigner, Research and Development Engineer
Meero, the largest photography company in the world today, has built an image editing AI capable of editing photos automatically in just 60 seconds (compared to four hours for a professional editor).
To get a better understanding of what that actually entails we’ve spoken to Juliette Chataigner, Research and Development Engineer at Meero, who has explained to us how these revolutionary algorithms work.
Can you tell us about yourself, your job, and tell us more about how Meero Artificial Intelligence works?
I’ve been working at Meero as a research and development engineer for more than a year now. I studied photonics (light sciences), and I specialized in photography and digital image processing.
At Meero, we work a lot on image processing using so-called artificial intelligence methods.
In the collective imagination, AI is a computer that learns how to perform manipulations on its own from data. But that's not always how it works.
We have AI algorithms but we also create more "typical" image processing algorithms, which consist in performing mathematical operations on pixels without any notion of artificial intelligence strictly speaking.
Can you briefly explain how these algorithms are able to edit images?
Well, to start, you need to know that the images we are working are really matrixes with three RGB layers (Red, Green, Blue) superimposed, one on top of the other. Each triplet of values builds the color of the pixel associated with the image. All our work is based on this: we work with these values, modifying or analyzing them according to what needs to be done. We’re then able to modify the colors, detect or even outline the contours, and many other things.
Here are some examples of photo editing done by an AI algorithm that we’ve implemented at Meero. This intelligence has learned how to "enhance" a photo automatically, according to specific criteria based on manual editing by professionals.
Let me explain: at Meero, we have a database with thousands of images that have already been edited by editors and graphic designers. We teach our algorithms to do the same editing following this example. In the end, the algorithms are able to do it on their own by reproducing the patterns seen in the learning base.
How do you plan to improve Meero's AI, and how far do you think it could go in the near future?
Meero's AI is of course constantly being improved.
In order to meet the new customer requirements and to face the competition of this market, we are constantly developing new projects and perfecting our existing algorithms to increase our performance.
At the rate at which Meero is evolving, we’ve worked so far mostly on short-term projects, aimed at meeting the immediate needs of photographic editing. Now that the team is well structured and the first phase is over, we can focus on longer and more complex projects. This was the case, for example, of our algorithm that changes grey skies into blue skies, which was our first real long-term project.
Our goal today is to create a true research entity that will allow us to go even further in our approach to photography. This is what we are starting to do on subjects like the evaluation of the aesthetic of an image, or generating photographs automatically.
Can you imagine Meero completely generating images in the future?
On the research side, I think so.
Generating images means creating algorithms capable of producing images themselves, from some pieces of data, sketches, or texts, for example.
At Meero, we have a Ph.D. candidate who is beginning to write a thesis on the subject, and on GAN (Generative Adversarial Network) technology. It's a topic that is very different from what Meero is doing today but could have other applications and allow the company to develop in other technical areas. I don't have that kind of insight, but, after all, isn't it the goal of any Tech company to continue to innovate on ever-widening themes?
What about the R&D team?
Today we have a team of about fifteen engineers with a wide range of profiles, image specialists, IT specialists, doctors or engineers from different fields.
Thank you for the interview and all this information Juliette!
Curious to learn more about Meero’s technology? Stop by the technology web page, and be sure to follow Meero on Twitter and Linkedin as we post regularly about the events and tech conferences where we will be speaking.