Luca Boschin, CEO, VISUA: The complex and diverse world of Visual-AI
AI News sat down with Luca Boschin, CEO and co-founder of Visual-AI solutions firm VISUA, to discuss the growth of the company’s offering in recent years and the latest trends in visual artificial intelligence.
AI News: What unique solutions do VISUA bring to the AI industry?
Luca Boschin: VISUA has applied Visual-AI (also known as computer vision or vision AI) to numerous use cases since our inception in 2016. This started with brand monitoring, where we process hundreds of millions of images per month along with tens of thousands of hours of video to find brands mentioned visually, be that through a logo in an image or a brand name appearing in a video.
We also combine this with object and scene detection and visual search to extract key visual signals. For instance, it’s one thing knowing that Budweiser appears in 500,000 images in a month, but what is really critical to know is where Budweiser shows up. How often is it next to food? How often is it with football on the TV in the background? Does Corona show up more outdoors than indoors? This kind of data is really useful for marketers.
Recently however, we have adapted our tech stack for highly specific tasks, like sponsorship monitoring in live video feeds, counterfeit product detection, copyright infringement detection, and digital piracy monitoring. Most recent of all, we’ve added visual authentication of holograms and the detection of graphical attack vectors in phishing attacks.
In each of these use cases we look for what we call ‘visual signals’. This is the important unstructured data that is locked in visual media. Our Visual-AI can extract that data and report on it, delivering the insights required for each specific use case.
AN: What are some of the latest developments at VISUA?
LB: We recently added holographic authentication to our offering in partnership with De La Rue. Holograms have really revolutionised the world of brand protection because they allow brands to inexpensively provide a visual cue of their authenticity. But perhaps because of their popularity, bad actors started to create fake holograms to go with the fake products. These fake holograms were virtually indistinguishable to the naked eye without specific training or a genuine comparison. De La Rue, a key leader in the area of hologram labelling, needed a way to solve this and having reviewed many different offerings, chose VISUA to help them deliver a solution for quickly and automatically authenticating holograms. Just point a smartphone at the hologram and it will tell you if the product is genuine or fake within a few seconds.
Secondly, we’re really proud of the work we’ve done in cyber security, and particularly phishing detection. It’s amazing that bad actors are also using AI. But they use it to make detection difficult. Most recently they’re also using graphics to confuse victims and hiding trigger words. That makes these elements really difficult to catch. So platform providers, managed detection & response companies, and threat intelligence services all need more data and early warning systems to allow them to quarantine suspicious emails and websites for deeper analysis. Our Visual-AI provides that to them.
AN: What are VISUA’s plans for the coming year?
LB: We’re working really hard in the cyber security space. There is a great deal of interest in tackling this growing issue of graphical attack vectors. So, we’re working with various companies in this space to help them detect and block malicious content more effectively. Meanwhile, we are looking at other possible opportunities and verticals to see which are the most viable and worthwhile to pursue. I am sure that other AI providers feels the same way, but our problem is not one of finding opportunities, rather it’s identifying which are the best opportunities to pursue among the many that present themselves to us.
AN: What trends are VISUA noticing in the AI industry?
LB: Too often we see companies underestimating the complexity of computer vision. Companies like Microsoft, Google and Amazon offer APIs that allow you to access impressive computer vision technologies. But in most cases these are either limited in their ability to be adapted or require extensive knowledge to implement. They might look ‘off-the-shelf’ but in reality, they’re not even Lego blocks; they’re the plans that allow you to mould the Lego blocks to build your system.
We took a decision from the start to not be an API company. There’s a good reason for that. It’s relatively easy to build a prototype using off-the-shelf solutions. It gets shown the board and everyone claps and says, ‘OK, let’s create the full production version’. That’s when things go wrong because scaling computer vision so that it’s accurate, efficient and cost-effective is really hard! Several clients tried doing it themselves. A year later, with lots of wasted budget and lost opportunity, they admitted defeat and approached us for help. Within weeks they were operational!
API companies tend not to offer good support. If you need help you pay a lot of money for an extra support pack or you hire in a consultancy firm to help. Even then, if you haven’t set the parameters and brief for the project, the wheels can come off really fast. We saw this issue early on and we kind of bucked the trend. We saw that companies wanted to implement Visual-AI, but their brief needed padding out. Sometimes they didn’t even know what questions to ask themselves to develop their brief. That’s where we come in. We are the Visual-AI experts and can help guide these projects to success. We’re not consultants, and that’s not the main focus of our engagement, but we recognised that that’s what companies needed, just as much as access to our API.
AN: What does VISUA plan to discuss at TechEx Global?
LB: We’re participating in the Cyber Security & Cloud Expo and our marketing director, Franco De Bonis, will be talking about the growing threat of graphical attack vectors and how Visual-AI can help mitigate or even eliminate that threat. But although we’re there for the cyber security event, we love discussing all things computer vision and we love a challenge. So if someone reading this has a particularly gnarly project that they think Visual-AI could fix, come and find us!