The project
Arlo (a brand of Netgear, Inc in 2014–2018), an ITS’s partner, needed to develop and launch a service module capable of analyzing short video clips and confirming or denying the presence of people in the frame. Today an average middle developer will solve such a problem within a couple of days. Back then, in 2015, six months before the Caffe project (the first open-source image classifier framework) emerged, it was a real challenge.
Today we can and do much more and faster. Arlo home video surveillance systems can inform the client about the appearance of people, cars, and various animals in the field of view, detect mail left at the door, and recognize baby cry (Arlo nanny). We are working hard to improve the identification of people, cars, and animals. It matters indeed who is at your door, who drove up to the house, and what animal is walking in the backyard: someone else’s dog or your cat on a spree.
Of course, the world evolves and all these problems already have a solution in some form. However, I would like to point out: We are talking about business, competition, and the quality and cost of the services sold. Our service processes one million requests per day, and we cannot mess up either by indicating the person in the frame as a dog or not detecting a potential robber. Therefore, we are constantly working to improve the accuracy of our algorithms, studying and trying to apply recent developments in AI. We are also looking for ways to reduce the service cost: how to reduce the required computing resources and increase the bandwidth while maintaining quality.
The team
The guys from our team are constantly developing. They learn on site by doing. This is a big plus of the project.
Of course, after so many years of work on a project in such a new area, we faced the need for building AI-specific development processes. Work processes are not new in the software development industry. However, there are many significant nuances in the AI area. I’ll save you from examples and will only mention that the ML CI/CD (Machine Learning Continuous Integration/Continuous Development) term appeared only six months ago.
Everything is yet to come
Artificial Intelligence is an area that is now going through its post-puberty period. A combination of a massive increase in capabilities and a cost reduction, as well as an enormous variation in all kinds of applications, is still ahead. All you need is hands, a bright mind, and eagerness: take it and do it. There is enough work for everyone! ITS was extremely lucky to have such a project in 2015. And I was very lucky to be in ITS during this period, to be involved from the project’s very inception, and to be leading it up to now.