Dr. Radu Rusu from Fyusion talks about how they combine 3D imaging with deep artificial intelligence.
1. Tell us how you came to be the co-founder and CEO at Fyusion. What does your role as the CEO demand of you on an everyday basis?
I come from the mountains of Transylvania, Romania, and had a long detour through Germany before I landed in California about ten years ago. It’s been an exciting journey and I had the pleasure to meet a lot of smart and humble people along the way that shaped the way I function and think, for which I am very grateful. As someone that liked to play games and code at a young age, I was the exact opposite of my father, an economist, who dealt with issues related to business development and money/numbers. We used to laugh that I’d never care about such issues, and that I’d be a geek forever.
From being an engineer, I turned into a researcher, and then into a founder/CEO. Funny how things turn around, and I guess the joke is on me. Starting a company typically comes from a pain point, and that’s what I experienced as well. Fyusion was founded because of a need that we felt wasn’t being addressed: we got to a point in our society where the impact of solving problems through the use of Artificial Intelligence and visual imagery is profound, yet we have been limiting ourselves in providing the best possible solutions due to technicalities such as leveraging subpar visual data — i.e., 2D images.
My role on a daily basis changed based on the stage Fyusion is in. I went from being focused 100% on development in the early days, to building a team, raising funding, managing, business development, and wearing many other hats based on where I was needed.
2. How did you choose 3D technology to be your career path? When and how did the seed for Fyusion to be your tech venture get sown?
We should not rely on 2D images to build machine learning models to solve critical problems in our society, but rather identify the problems first and then find ways to solve them efficiently and accurately.This is not a new concept, and fields such as robotics have been dealing with it for a very long time, where teams tasked to solve a particular hard problem will go and build complete hardware and software systems from scratch in order to achieve their goals. In autonomous driving, we’re not duct taping a smartphone to a car while driving, but rather instrumenting the car with a multitude of LIDAR and camera sensors. In massive industrial applications, we’re building robotic arms and complex laser-line and time-of-flight cameras that create incredibly accurate 3D models of objects and use sophisticated computer vision and machine learning algorithms to solve hard problems. So why is it that for everything else around us we don’t apply the same approach?
This is what Fyusion set out to do: identify the needs of a new visual format that is 3D and spatial, create the underlying platform, algorithms, technologies, and find the right use cases for a new visual format to exist and thrive. We coined the new format “.fyuse” and it sits alongside a “photo,” which is a flat 2D projection of the world onto an array of pixels, and a “video” which is a series of “photos” recorded at a certain speed. Our “.fyuse” format records spatial information, renders it using light field technologies, and is able to extract much more accurate information such as tiny damages on a vehicle in an inspection application. Most importantly, we do all this with software, and use commodity cameras and off-the-shelf hardware such as smartphones, thus making our whole approach incredibly scalable.
3. What are some of the business use cases for the use of Fyusion software for enterprises? Share some typical pain points your customers come to you with and the ROI Fyusion scores against its competitors?
We compete with different companies in each of our product areas. Our products typically take old-school 2D functions and deliver them in 3D in areas where there’s a clear benefit to the consumer, such as automotive inspections or different types of e-commerce imaging. Our competitors are usually older companies who are struggling to catch up. More often than not, we’re competing against the status quo.
Like society as a whole, our customers face an extraordinary amount of change and don’t have the bandwidth, so they need solutions like ours that make it as easy as possible for them to modernize, by providing something more interactive than an old-fashioned photo. . From the founding of our company, we’ve been all about innovation in computer vision—we’ve compiled over 150 patents to support our products. But in 2020 we became a customer service company that happens to deliver innovative computer vision products. As we’ve gained EQ as individuals, we’ve also gained EQ as a company.
4. Vehicle inspections using computer vision 3D augmentation can be a great assist for rentals, auctions, dealer leasing, or insurance purpose. Can you elaborate on how this kind of automation can help the automotive industry?
We see the entire automotive industry undergoing a major digital transformation, beyond the basic “infotainment” systems that until recently were the only digital components in your vehicle (apart from Teslas and other electric vehicles). This is manifesting itself in autonomous driving technologies on one end, and then in the process of capturing, understanding and visualizing vehicles and any information about them on the other. The latter pain point is really where Fyusion is helping to revolutionize the entire process by leveraging modern Computer Vision and Machine Learning technologies to bring in accuracy and consistency via automation, which ultimately translates to an enhanced experience and most importantly, trust. Instead of relying on an inaccurate and inconsistent human analysis of a car’s condition, Fyusion brings in systems that leverage camera data and Artificial Intelligence to automatically understand and inspect the vehicle. As you noted, this is applies to any sub vertical including logistics, wholesale auctions, car-sharing and rentals, insurance, remarketing, trade-ins, and many more. Effectively we give the power of an automotive expert that’s never tried to any person that can move around the car while holding a smartphone running Fyusion software.
5. Can this also not be extended to the real estate industry for assessing commercial properties for enterprises?
The applications of our technology platform are numerous and include that as well. However, as a small company, we learned that focus is very important to solving deep problems that your customers care about in a specific vertical, so we’re definitely concentrating our energy on automotive first. As we mature as a company, we will expand to several other areas where we see the value add both for us as a business as well as for our customers.
6. How can healthcare and other industries that work in hazardous conditions, say mining, and benefit from computer vision-based 3D imaging and augmentation? Can it be a prescriptive tool where it helps decide the safer next steps or choices for surgery or entering dangerous terrains?
Healthcare is dear to our hearts as we should all strive to improve the way we live and our bodies function this century, in order to increase our individual lifespans and be able to accomplish more as a species. In this regard, technologies such as what Fyusion developed can be applied very rapidly to any problem that can be monitored easily using camera systems on a regular basis. For example, just as we identify dents and scratches on a vehicle, we should be able to identify skin conditions and irregularities that might lead to cancer or other skin diseases in the future. By hoping that one day we can partner up with leading medical companies in this field, we could ultimately provide doctors with a complete “skin history” of a particular part of your body that you’re worried about and answer questions that a doctor simply cannot just through your regularly scheduled visits, like for example: “has this mole been growing over the last X months?” which should help diagnosis and provide visual introspection for those patients that are prone to later suffer from such diseases and prevent problems from happening before it’s too late — not to mention that getting access to the best doctors can only be achieved through digital technologies in the future, as there won’t be any scalable way for everyone in any remote corner of the planet to see the best physicians in person. This is obviously very high level and will require a lot of other issues to be solved concurrently, but as a technology company we are working on these building blocks as we speak and would love to improve our quality of life in these areas.
7. Could you share some of the upcoming trends that can be expected in the AI/AR 3D imaging domain? What industries are expected to most benefit from these advancements?
We are mostly interested in the evolution of the “data formats” at Fyusion, meaning that we have full confidence in the advancements in hardware performance and miniaturization separately, and we believe that from an algorithmic perspective we already have most if not all of the ingredients that we need to solve really difficult problems. Where we really are lacking is the data problem: we have built an entire Internet filled with visual data for completely different purposes, and attempting to leverage that data (e.g., 2D photos and videos) to extract high accuracy answers for serious problems has failed us. It is no surprise that fields such as robotics (including self-driving cars) have been looking into 3D data for a very long time as the only answer to this issue. This is where Fyusion, given our previous experience and pedigree in robotics, is hoping to do the same. We want to bring together a plethora of technologies with novel 3D data representation in ways that are fully scalable, and can reimagine the way we capture, visualize and leverage this data with Machine Learning algorithms to bring “robotics accuracy” to solve problems in other fields. As an extra added benefit to representing the data in 3D, we can now also obtain a digitization of our surrounding world that can work natively with AR/VR/MR applications, unlike 2D which will always be a flat subpar representation in those mediums.
8. Can you give us a sneak-peek into Fyusion’s upcoming upgrades or new features?
We are hard at work to continuously improve our light field capture technology and processing, to allow more and more people on any smartphone or other camera systems to capture high quality photorealistic 3D data with ease. There’s a number of improvements across the board in our algorithmic pipelines that are being released in a regular fashion, and we are grateful for all the advancements in compute performance that we are making as an industry. That allows us to simply do more on the same edge devices that we have in our pockets.
9. What AI innovation you will be on the lookout for in the upcoming years?
The number one thing that we’d like the industry to solve from an efficiency perspective is the “bandwidth” issue. Perhaps this topic should be handled by smarter and smarter systems, because the requirements for being constantly online are only going to grow as we develop more sophisticated applications in various fields. Connectivity options like 5G are very well regarded in this sense, but the fear is that with higher resolution video streaming and other gaming they will simply get saturated very fast, thus allowing little room for improvement without throttling — which should never be the answer. Instead, we hope that more Machine Learning based approaches can be leveraged to create a better balance between such complex interaction models, where extremely large numbers of nodes all get connected through the same networking hubs. In my opinion, the more we solve the bandwidth issue, the more we’ll be able to advance everyone’s work in any field, as the duality of edge and cloud computing can now work more seamlessly together to accomplish our goals, rather than boxing ourselves into the current generation of smartphone hardware.
10. How do you think the upcoming AI professionals can be nurtured to think out-of-the-box to resolve higher global goals with responsible AI?
That is a very tough question, and I think the simplest answer I can give is: through education. Like anything new, things are fairly unknown to most people and unregulated from an industry perspective. We need to give ourselves time and apply education and information sharing principles between those that have that expertise and allow humanity to look at AI for what it really is: just a tool, that when used properly can be extremely beneficial to our future. Any technology whether past or present can be leveraged to do both harm and good. We simply have to understand that and educate ourselves on where that line should be drawn, and we’ll be okay. I have full confidence this is going to happen soon and will come naturally.
11. As the CEO of a tech company, how do you inculcate a culture of innovation and problem-solving amongst your employees irrespective of the range?
Absolutely. The best ideas can come from anywhere, and we have a culture of healthy pushback and innovation that allows us to brainstorm across the board, in every aspect of what we do. There is simply no other way to “compress time” which I believe is one of the most important aspects of being in a start-up.
12. What is the one leadership motto you live by?
Dr Radu B. Rusu
Dr. Radu Rusu, co-founder and CEO at Fyusion, has a background in 3D imaging and robotics. President of Open Perception and a world-renowned expert in 3D data processing, Dr Radu Rusu has almost 20 years of experience in the field. Radu was named Top 100 Most Intriguing Entrepreneurs in 2016 by Goldman Sachs, and has been awarded the IEEE RAS Early Career Award in 2013 for his contributions to the field of 3D Computer Vision. He has won the Open Source Software (OSS) World Challenge in 2011 with PCL.