Dr Appu Shaji: Low-code computer vision: how humans are training the future of AI
How do you adjust quickly to the post-pandemic world? Software inevitably plays a role, but the rush to build and deploy systems for the new normal has put enormous pressure on internal and external developer resources.
Instead, many businesses are turning to ‘no-code’ software. As the name suggests, no-code does away with the need to learn programming languages. Instead, everyday staff can build apps using a drag and drop graphical interface, reducing the need for expensive developer resources. It’s a process that’s familiar to anyone who’s ever built their own website, although you are more likely to see it being used to build business reporting dashboards or to automate common business workflows such as employee onboarding.
But things really get interesting when no-code tools are powered by artificial intelligence. Rather than dig into the complexities of machine learning and call on the expertise of developers and data scientists, no-code concentrates on the most important elements that are understood by business users, adds a simple UX (user experience) and abstracts everything else.
From antique vases to sporting superheroes
Computer vision, which focuses on image recognition, classification and creation, is a prime example of how no-code can help organisations get greater value from their visual assets, be they photo archives, video libraries or marketing agencies. Simply put, no-code computer vision enables the subject matter expert to train an algorithm to recognise a particular category of image, from landscapes and buildings to historical figures and antique objects.
Take a museum that licenses photographs, or a football club with an archive that spans the decades. Using a simple interface, an employee can drag and drop a workflow that trains the algorithm to recognise a rare style of antique vase, or all the images of a recently deceased club hero.
The other advantage is that today’s computer vision algorithms are smart enough to work with a relatively small number of images. In the case of Mobius Labs, a Berlin based startup, the end user selects just 20 positive and incorrect images to train an algorithm. Once ready, it can then search the entire archive and identify every instance of a rare porcelain vase or a much-missed centre forward. This includes overlooked content where the subject might be tucked away in a corner or partially obscured.
Want AI to do better? Hire a human
Photo archives are also keen users of low-code computer vision software. Under pressure from everyone from Pinterest to ‘free’ stock libraries, many are turning to computer vision to ‘sweat’ greater value from their collections. There are many ways to approach this opportunity. For instance, you could train the software to find photographs of a historical figure who is back in the news. At the other end of the spectrum, you can quickly create collections that meet the current news cycle. For instance, searches for images associated with pandemics surged during the early days of the coronavirus crisis.
Looking to the future, Mobius Labs and others expect the audience for low-code computer vision to expand beyond professional photo collections. Over their lifetimes, most consumers will accumulate tens of thousands of photos and videos on their smartphones or stored on the cloud. The next generation of ‘DIY’ computer vision will enable individuals to search for personal content in their photos, not just the generic landscape or facial recognition classifiers available today.
As for businesses, it is hard to think of any industry that won’t benefit from employee-trained computer vision software. Right now, everyone from manufacturers to healthcare organisations rely on massive volumes of data to spot assembly line errors or diagnose a serious illness from a scan. Imagine if the expert in each case, an assembly line manager or a radiologist, could input their knowledge into the process. Rather than replacing employees, AI might increasingly rely on human experts for their training. If you want a job in the future, learning how to use computer vision low code tools sounds like a good place to start.
For further information visit www.mobiuslabs.com