The Apply AI podcast explores how Artificial Intelligence and Machine Learning can revolutionise businesses. Solutions that save time and money, increase revenue and future proof their business.

World’s Fastest Computer Vision AI: Xailient & Shivy Yohanandan

Apply AI speaks with Shivy Yohanandan from Xailient, a computer vision company that was born out of nine years of research around how humans and animals perceive the world. Xailient’s mission is to improve the efficiency and accuracy of computer vision in the modern world.

Listen below or on your favorite listening service: YouTube | Spotify | Apple

The following summary was generated by Sonnant‘s SEO ready content & keyword transcription. Sign up for a free trial.

Free Sign Up

What are the more interesting applications of computer vision technology?

There are many applications for computer vision AI and pretty much anything you can imagine a human doing with their vision – we can train AI to do something similar. For example,  take out all the noise from what you’re seeing and focus in on objects.  If we talk about computer vision at large, we’re talking about things from reading license plates to counting people to detecting cars to spotting sharks at bondi beach with drones to warn surfers. So it’s an infinite solution space, but Xailient solves one critical problem in that space.

And so with this huge, broad brush potential application, how do you identify target markets?

In terms of identifying a beachhead, we are looking particularly at access control. So access control where you’re using, for example, face authentication to let people into a premises through security surveillance. And then gradually we’ll start creeping into other verticals.

The technology behind computer vision AI, what does that look like?

It’s essentially teaching computers how to see like humans, given an image from a camera. You want the AI To detect certain things and then process those things. For example, retail stores want to use their existing cctv cameras to check if people are shoplifting. So you first want to detect the people and then you want to see what they’re doing with their hands. Get an algorithm to then process those images to then generate analytics and you’re then doing the retraining.

Our mission is to enable this future of pervasive, ubiquitous computer vision. We want allow developers to decide what they want to detect and do with images. We provide a platform and they can come with training data.

Let’s say someone wants to detect Varroa mite on bees. Like one of our partners. They would need to come with videos of bees with the Varroa mite. That gets fed into our machine online, which trains the AI to then recognize bees that are carrying mites.

How do you improve and train the AI model over time?

One of the biggest problems that’s been around for a while that has been stagnating the technology is the notion that you train one AI model once and then you set and forget.

That’s definitely not how it works and what we’re finding, dealing with a lot of customers who want to scale AI to many, many cameras. The idea of training one model and getting it to work on hundreds of cameras doesn’t work. You need to continuously fine tune and have that sort of feedback loop that retrains the over time so that it gets better over time just like humans – just like a child.

What really gives us the cutting edge advantage at the moment is that not everyone in computer science can claim that they understand something fundamental about the neuroscience of vision or the evolutionary origins of vision. That is the key to solving a lot of problems of computer vision today. To really understand why animals and humans can process information efficiently. That is our unique differentiator.

What are the primary motivations for a business to adopt Xailient?

I think a primary motivation in computer vision is money – being able use a high bandwidth of information with relatively cheap devices, like cameras and tiny computers. You’re able to then automate a lot of tasks, which would otherwise take up a lot of manual work with fatigue mistakes.

When you’ve got cameras rolled out in supermarkets, monitoring what people are doing with their hands, that significantly improves and reduces the amount of lost revenue because you can stop people from shoplifting. The other case would be lowering expenses for shopping centres and car parks – reading licence plates for ticket machines for example. Now you can just replace those machines with very cheap cameras hooked up to a Raspberry Pi with AI running on board, scanning every car that comes in and checking it against a database.

What investments are required to get started in implementing computer vision?

Up until now, most people who want to implement computer vision in their business needed to have a pretty high set up cost because a lot of the computer vision is processed on the cloud. You would need to stream all the video to a cloud machine and cloud computing is very expensive. The benefit with using Xailient is we can run on the camera itself – that way your unit economics are significantly improved.

Xailient has built the world’s smallest computer vision – smaller than any anyone else by a long shot, we can actually fit our software algorithm on the camera itself. The camera talks to the cloud and downloads the latest version of the algorithm. Just like everyone is used to for software updates on their mobile phones.