Every once in a while, a new product comes along that changes everything. Microsoft Windows; Google; iPhone; each one was groundbreaking, not necessarily in its technology, but in the way it brought that technology together and made it accessible to everyone.

The In-Sight 2800 Vision System from Cognex gets me more excited than anything else I’ve seen in factory automation in the past 30 years. It has the potential to kick-start the transformation of manufacturing, production, and industrial productivity like few things before it. 

We’ve been itching to talk about it ever since we first tested it. Here, in our latest deep dive, I’ll explain why.

What is it?

The 2800 is a new smart camera that can be used to automate visual inspections. It brings together deep learning technology and traditional automated vision inspection tools in an easy-to-use way to solve all kinds of application challenges. 

Designed for factory automation, it allows manufacturers to solve complex application challenges, increase product quality, maximise operational efficiency and customise hardware for specific application needs.

Product quality: Catch small, subtle defects with highly accurate, deep learning-based error detection capable of solving applications and classifying parts with variation based on multiple defect types or user-defined features. 

Maximise operational efficiency: Leverage your existing workforce and keep your lines running with fast, intuitive job setup—no vision or deep learning expertise required. 

Customise hardware for specific application needs: Modular solutions with integrated lights and lenses takes the guesswork out of image optimisation. Identifying the correct combination of accessories for your application is only a few button clicks away.

With a starting price of around £5,000 – it makes deep learning an affordable, cost-effective option for the majority of the market, for the first time. The technology inside the 2800 simply hasn’t been available at such an ‘entry level’ price before.

It can be bought as a stand alone camera, or with the option for automatic software upgrades in future.

A production line using the Cognex In-Sight 2800 technology.

What’s changed?

The 2800 doesn’t do anything we can’t already do using other, more advanced deep learning cameras like the D900. But it is cheaper, easier to use and in many cases just as effective.

Deep learning is shorthand for software capable of handling complex and highly subjective visual inspections that previously only humans could do. It does so by learning over time what counts as ‘good’ or ‘bad’ in a particular visual inspection.

Until now, deep learning has only been available to experts in machine vision. All other systems on the market, including the more advanced kit in the Cognex range, requires a high-level of computing and coding expertise to set up and use. So much so, that they’re only really available with expert operator training included.

The big change in the 2800 is the simplicity of the interface between the user and the software inside the camera. In other words, what’s on screen and how easy it is to use.

Think of the move from MS DOS to Microsoft Windows, or, the move from older ‘dumb’ phones to IOS and the App Store. One of the reasons why smartphones became so popular is the simplicity of their app-based interface. And it’s the same for the 2800.

The technology inside is not new – deep learning has been around at Cognex for at least three years – but now anyone can use it. While the computing power stays exactly the same, almost anyone can master it in minutes, with a clear screen to follow, big simple buttons to press and fewer detailed options to choose between.

Integrated lighting showing the red, blue, white and green light settings.

Who is it for?

A limited number of options will reduce the chances of getting things wrong for the less experienced operator, but it does mean less flexibility when compared to the more advanced cameras.

However, unless you have a highly sophisticated application to inspect, the In-Sight 2800 will fit the bill.

The 2800 is for people who have a problem they want to solve through visual inspection. Classic vision technology can handle big, obvious issues, such as whether a product is physically there, or not. But when things get more complicated, with more possible permutations, that’s where the 2800 steps in.

Possible applications include categorisation, quality checks on fast moving consumer goods, pack checks for sealing issues or defect inspection on cables, gears or gaskets, for example. Anywhere that classic vision technologies would struggle.

It could be a factory that’s having quality issues – for example, seemingly random defects on a production line where understanding the nature and scale of the issue requires inspection of every item.

Or it could be a factory where pack seals need checking for quality, but the shiny, transparent appearance and material properties make the job of checking for errors hard even for humans. 

Classical vision technology can’t handle that level of possible variation, but deep learning can.

There are still countless examples of companies using unreliable manual checks for routine inspections. The 2800 can automate those inspections and solve application challenges that could previously not be automated, at an affordable price. 

The integrator’s view

Stuart Jackson, Bytronic

“We recently travelled to Germany to test it out and I was really impressed. Yes it has a lower price tag, but it costs less to integrate too, so it opens up a whole new market. It’s a nice piece of kit – especially the integrated lighting package which lets you swap between red, green, blue and white light to suit your applications.” 

“I’m looking forward to testing it out for an automotive company that wants to run an absence test on bolts on a gearbox production line. Lots of variables in there – depending on the orientation of the bolt, and whether it’s oily or not, they can be hard to detect – so this could be a good solution.”

How does the interface work?

All vision applications consist of the same five steps.

  1. Image
  2. Inspect
  3. Communication
  4. Inputs/ Outputs
  5. HMI


The first step is always to get an appropriate image for the inspection – A sharp image, that’s well lit with the right colours, and so on. On the 2800, brightness and focus can be optimised through just two options, compared to dozens of options previously.


The Inspect Tool is where the real inspection takes place. On the 2800 interface, the tools are clearly displayed, with just a few simple options to choose from.

The final three steps, communication, inputs/outputs and HMI are all about analysing and visualising the image data. With the 2800, they’re all done on the same intuitive interface.

Cognex Q&A: with Mathias Maas

What’s next for vision?

“There are still lots of deep learning tools available on our most advanced cameras that will be rolled out to the 2800 in the months to come. Tools such as OCR – Optical Character Recognition – for reading even very challenging text with a high degree of accuracy. There’s a good chance that the next tool we release for 2800 will be OCR.”

“We have hundreds of developers working on these tools to bring out more deep learning-based tools that are easy-to-use, easily implemented and easy to maintain. Beyond that, we can’t give any more details on the next big advances right now, but we have teams of engineers working on the next big advances in deep learning that are coming further down the road.”

Why does Cognex choose to partner with Bytronic?

“No one does deep learning like Cognex – and Cognex is great at developing new products, vision algorithms, tools and hardware, but the Bytronic team brings that technology to life. 

“They know what the customers need. They have specialist engineers who can integrate these products to solve real-world inspection problems. By working together, it gives the customer the best of both worlds. They get the best products, and a solution that solves their problem.”

In depth: In-sight 2800 Vision System

Intuitive development environment simplifies setup : With its point-and-click programming, the EasyBuilder® interface guides developers step-by-step through setting up the In-Sight 2800. From image capture to the final result and beyond, both new and experienced users are able to configure vision applications.

Innovative edge learning tools automate tasks, fast: The In-Sight 2800 is embedded with powerful ViDi EL tools that leverage a pre-trained set of deep learning algorithms to solve complex applications, like advanced segmentation and classification, in minutes. Training is simple, requiring as few as five to ten images per class and delivering reliable results in real-time.

Advanced multi-class functionality provides ultimate flexibility: Go beyond the standard OK/NG output with the ViDi EL Classifier tool. The tool identifies and sorts parts based on multiple features or characteristics. This enables users to classify defects into different categories and correctly identify parts with variation, allowing them to automate a wider range of tasks.

Bring us your challenge

Whatever you’re trying to achieve, let our vision experts help. Speak to the team.