Heiko Eisele, president of MVTec, LLC speaks with Winn on the set of Manufacturing Matters at the 2022 Vision Show in Boston. MVTech is a well-known international producer of machine vision software. By supplying cutting-edge technologies like 3D vision, deep learning, and embedded vision, MVTec products enable innovative automation solutions for the Industrial Internet of Things and are employed in all demanding imaging applications.

Winn Hardin: [00:00:00] Thanks for joining us today here on Manufacturing Matters. We’re in Boston at the Autonomous Mobile Robotics and Logistics Show, co-located with the Vision Show in Boston. And I’m lucky enough to be here with Heiko Eisele, president of MVTec, one of the leading software developers in the machine vision space. Heiko, thanks for joining us today.

Heiko Eisele: [00:00:14] My pleasure.

Winn Hardin: [00:00:15] Fantastic. So, first of all, what’s the first impression of the show? What are you seeing that’s cool?

Heiko Eisele: [00:00:21] Well, first of all, we’re excited to be back in Boston. I think the last time was in 2018. And I have to say, I really missed it.

Winn Hardin: [00:00:26] Yes. Boston is a great town.

Heiko Eisele: [00:00:29] Well, of course, we live here. That helps very much enjoy. Fantastic being back here at the Vision Show.

Winn Hardin: [00:00:36] Fantastic. So what is MVTec focused on in terms of the applications of vertical industries here at the show?

Heiko Eisele: [00:00:46] Well, as you know, most of all we’re trying to solve customers’ applications. And everyone is talking about deep learning. For us, deep learning is really a tool. We see it’s important. It’s a great enabler. It allows us to do things that we haven’t been able to do before. However, to us, it’s just one of all of the many tools. As far as global industries, we do see applications, for example, in automotive to do various types of inspection for defects, also logistics, certainly a big growth industry and big potential. And dictation, label reading, OCR, barcode/data code, robot picking, of course, which is another application where deep learning can be applied. So I think the time is right in the years to come.

Winn Hardin: [00:01:37] So let’s talk a little bit, let’s delve into that a little bit. You mentioned deep learning. HALCON, your premier flagship product at MVTec, has both a traditional IDE, has traditional machine vision algorithms built in, one of the most powerful softwares out there on the market today. And then it’s fully integrated with deep learning capability, right? So how does that help you to better respond to the customer needs that are out there?

Heiko Eisele: [00:02:00] Well, what we see is that quite often deep learning doesn’t solve the entire application. Deep learning only is part of the application. Often you need some sort of preprocessing. You need to maybe just do some part party and reorientation environment [unclear what he’s saying here]. And this is where we think it is really helpful for our customers to have all the tools integrated into one package that can be seamlessly deployed.

Winn Hardin: [00:02:24] So when it comes to deep learning, what do you say, Heiko, to the folks who are saying, we were expecting more. We were expecting deep learning to solve more. Do you do you think deep learning is on a good adoption curve? Do you think it’s going to be able to deliver on the full promise that so many people have? Or do you think that people are asking for too much?

Heiko Eisele: [00:02:40] To some extent, I believe that’s certainly true. I mean, expectations are too high. And I think at this point we’re a little bit beyond the hype curve already. I think we’re at a phase where the realization settles in that it can’t do everything. And that it’s simply not the right tool for certain applications. One of the big downsides of deep learning is the lack of adjustability. I mean it’s still a black box. And people on the production floor, they want to be able to fine-tune things and tweak things, and they can try it. And we still don’t fully understand how the technology works. Someone does magic, you know, using a lot of training images. But often there’s the sort of lack of accountability that bothers people still.

Winn Hardin: [00:03:23] So one of the great challenges I think for machine vision systems and software solutions is to be able to adapt to new product variations that might be very similar. So is there a certain rule or guidance that you give your customers that says, “Okay, we have a deep learning model. It’s optimized for your existing product base. You can expand this to a certain amount.” But what kind of guidance do you give them on being able to either expand the model, with new variations, or how do you address that? Or do you just say, listen, if we’re going to have a new product, we’re probably going to need a whole new training set and we’re going to start from scratch.

Heiko Eisele: [00:03:59] Well generally we certainly make customers aware that the model has to be trained with the stuff that it has to see later. So in other words, if it has seen stuff that it hasn’t been trained for, you could be potentially in trouble.

Winn Hardin: [00:04:12] It doesn’t have imagination.

Heiko Eisele: [00:04:14] Exactly. Exactly. So deep learning is, we always speak about it, we also refer to it as artificial intelligence. And in my opinion, it’s not really. There’s no intelligence behind it. I mean, in the end it’s just a statistical model of the training data that you’re building.

Winn Hardin: [00:04:29] All right. So when we think of MVTec, we think of HALCON, we think of the other software products. We think of traditional machine vision applications in the warehouse, in the factory. We mentioned logistics. But earlier during our conversation, we were talking about the future of agriculture, which really lies outside, in every conceivable sense of the word. So what does MVTec see about agriculture? What is the current state of the art in terms of using imaging systems and where are we going?

Heiko Eisele: [00:04:56] Well, first of all, I think in the agricultural sector we have sort of macro trends that simply drive a need for automation. We have a lack of labor on the one side and lack of, simply, resources, agricultural resources. That means we have to make use of our agricultural resources more efficiently. And machine vision automation can be a big enabler here. There are environmental issues like reducing the amount of pesticides and even water resources for growing. And this is all an area where machine vision can and will play an important role.

Winn Hardin: [00:05:32] Fantastic. Are we just getting into that agricultural growth curve now?

Heiko Eisele: [00:05:36] I would say we’re pretty early on here. You see, interestingly, a lot of customers that have specialized in that area have been offering a vertical solution, both hardware and software. But yes, as far as deploying the technology in the field, I believe we are pretty much at the early stage of the growth curve.

Winn Hardin: [00:05:54] Because these are large OEMs, there’s probably going to be a lot of specialized imaging systems, machine vision systems, either for weed detection, for example, or guarding a harvest or identifying what are good plants, how to best use water, and all that. Are these going to be new players that MVTec will be servicing or will MVTec be interfacing directly with the John Deeres, with the other large . . .

Heiko Eisele: [00:06:18] I would say both. I mean, you have big vertically integrated companies like John Deere who are developing their own technology in-house. But interestingly, you also see a lot of startups in that field. I mean, really small companies that just do one specific thing, maybe mapping fields to optimize irrigation or something. Or looking for weeds. So a lot of particular software companies are coming up with vertical solutions here.

Winn Hardin: [00:06:45] How does MVTec, which is an incredibly powerful software environment, but it’s also complex. I mean, to be able to do advanced solutions requires advanced software.

Heiko Eisele: [00:06:56] That’s correct.

Winn Hardin: [00:06:57] When you’re helping a customer develop an embedded solution, how do you shrink the footprint of that computational requirement? Is that something that MVTec is looking at or addressing?

Heiko Eisele: [00:07:08] Well, certainly, the idea always with software is the simple application should be simple to solve. For non-programmers, for people with not so much experience of vision. But there are these sort of more difficult applications where really you have to get that little extra juice out of the application as far as performance, as far as I can see. And this is where we can help with our more powerful tools and also with our application expertise that we make available to our customers.

Winn Hardin: [00:07:32] Do you provide a lot of consultation in that area?

Heiko Eisele: [00:07:35] Absolutely. Absolutely. We work very closely with our customers to actually make sure that they are using our tools in an optimal way.

Winn Hardin: [00:07:41] Fantastic. Here in Boston — let’s come back to where we’re at right now — what’s the coolest thing that you’ve seen on the show floor today?

Heiko Eisele: [00:07:48] Honestly, I haven’t had the chance to walk around much because thankfully there was a lot of traffic yesterday. So I spent most of the time talking to customers at the booth. But I hope maybe later today or tomorrow to check on some of the other exhibits.

Winn Hardin: [00:08:01] That’s the burden of being the president I guess.

Heiko Eisele: [00:08:03] Well, it is, but it’s a good burden to have.

Winn Hardin: [00:08:06] Heiko Eisele from MVTec, thank you so much for giving us some time today on Manufacturing Matters. If you’ve got any questions for Heiko, please reach out to us at Tech B2B. Feel free to share and like this video if you can. And until we meet next time, keep sending those ideas and tell us why manufacturing continues to be a very important matter here in North America and across the world.

Heiko Eisele: [00:08:25] Thank you.