How an Intel incubator helps start-ups get to grips with computer vision

10 Jun 2020

Jonathan Byrne of Movidius. Image: Intel

Jonathan Byrne, senior software developer at Movidius, tells us about Intel’s Edge AI incubator.

Back in May, we looked at seven start-ups participating in this year’s Intel Edge AI incubator, which sets out to help companies integrate computer vision into their tech solutions.

Now in its second year, the incubator was first held in 2019 in association with Dublin City University’s Talent Garden. Due to the coronavirus pandemic, this year’s programme has been held online.

Jonathan Byrne is a senior software developer at computer vision firm Movidius, which is now part of Intel. He spoke to us about his team’s role in the incubator and the challenges that come with running a highly technical incubator remotely.

‘It’s really interesting that these small companies are so nimble and can adapt so quickly’

Let’s start with the basics. What is the main goal of the incubator?

We essentially want to get people moving into deep learning or, in particular, deep learning on the edge, using Movidius technology. Movidius technology is designed to be low power and it’s not just for deep learning, but also for computer vision.

It’s not the easiest technology to use straight out of the box. There’s a lot of companies that recognise a business opportunity with the technology, but they might not know enough about deep learning to take advantage of that.

Essentially, the incubator is quite altruistic, from the point of view that we’re not taking a share of the companies or anything like that. We’re running a training course to help companies get a dataset, label the data, train a network on it, get the network running and to use the network to solve a problem, all while running it on Movidius.

What are the supports that Intel and Movidius provide?

Lots of technical support. Sometimes teams get lost in what they’re doing or feel like they’re not making progress. We let the teams try it all out themselves and if they succeed, happy days. If they don’t, my guys are able to say, “Well actually, if you try this setting…” There’s lots of stuff you only really pick up after years of doing this, so a lot of it is hand-holding.

It’s one thing to train a network and get it running on the neural running compute stick, then we’re like let’s move it onto the Raspberry Pi. They have a vision of what they want and we help them work towards that. It’s a mixture of project management and supports. A lot of the time you can train algorithms and they just don’t work. The problem could be a really small setting you’re only familiar with if you’ve been working with the technology for five years.

Now that the programme is being held online, how do you think it will differ to last year’s incubator?

One of the nice things about last year’s incubator, which was held in Talent Garden Dublin, was that there was a great network effect. There were people in there talking to each other, having the same problems and being able to help each other. That was really invaluable.

For example, one team wanted to learn about Docker, then all of a sudden the other teams wanted to hear about it too. So one of the guys who’s an expert on it gave a presentation the following week. It was that sort of thing that made the shared working space really nice.

Now, things are tough. A big company with loads in reserves can weather the storm just fine, but for smaller companies, it’s harder to find their way through Covid-19. What we did was move it online, it’s not the same, but we’ve been keeping up progress. We do weekly meetings, so it’s like online project management.

How do you maintain the social aspect of an incubator while holding it remotely?

We meet with the teams, we talk to them and see what kind of problems they’re having. We have a WhatsApp group where they can post questions and reach out to us with any issues they’re having as they’re moving along with the technology. It’s become slightly more distant but we’re still making progress by building a bit of structure and doing things week by week.

We also do paper reading groups and show and tell, where teams can show their progress to each other. What we want to offer is support. Any stuff they’ve needed – for example, we’ve had to ship hardware out to all of the teams – we help with that. We also help them gather datasets.

We’ve all had to adapt and if the teams haven’t got hardware they need, we’ve been able to source it for them so they can keep up development. It’s a slower pace. We’d be doing this a lot faster if we were all in the same office, but we are all managing.

While the current circumstances are challenging, do they present any unique learning opportunities?

It’s really interesting that these small companies are so nimble and can adapt so quickly. The quintessential example is Akara. They had a robot they’ve been developing for years, Stevie. One of the use cases in the old folk’s home was sterilisation, where they were using it with a UV light. With the coronavirus, they said: “Let’s just scrap all of that for now and just focus on a robot that can clean.”

They did that in the course of a week. Another example is Reivr, which was focused on asset tracking and medical tracking. They’re still doing that but they started to include databases so they have a backend that allows for tracking patients around the hospital.

Start-ups are very good at listening to their customers. Reivr went into Blackrock Clinic and asked if they still needed asset tracking, but found that it wasn’t the highest priority. Instead, the hospital was having trouble tracking patients so they quickly adapted their product to tailor to that.

Kelly Earley was a journalist with Silicon Republic