In my prior article for Fast Company Executive Board, I discussed how open-source AI tools are enabling businesses of every size to solve previously unsolvable business challenges without the need for huge AI budgets. Too often, media coverage of AI and machine learning (ML) only focuses on how the largest tech companies in the world are competing against one another to push the leading edge of the technology forward.
But that overlooks what I consider something equally significant: how smaller organizations can take advantage of the billions of dollars of investment in AI/ML that tech giants and universities have made over the past three decades, putting free open-source AI tools to work solving pressing business needs.
I gave a number of examples in my last column, including use cases for open-source visual identification engines and generative AI. Those examples were probably familiar to readers because they involve AI-driven technology that is already available in popular consumer apps and business applications.
This time, however, I’m getting a bit nerdier by talking about technology that you have probably never heard of: Gaussian splatting. That name is not a typo that spellcheck failed to fix. It uses mathematical techniques to create robust 3D visualizations in real time, which can be put to work in game-changing ways.
FROM VIDEO GAMES TO BUSINESS SOLUTIONS
Don’t worry if you haven’t heard the term before. You already know what it’s capable of, because Gaussian splatting powers the eye-popping visuals of your teenager’s favorite video games, the CGI-heavy scenes in new movies, and much more. The technology was developed in the 1990s to render cutting-edge visuals far more quickly without requiring heavy-duty computational workloads behind the scenes.
Gaussian splatting quickly established a foothold in the video game industry because it enabled game designers to create increasingly immersive visual worlds that could render and re-render in real time as gamers interacted with the environment. Using other methods, that kind of visual rendering simply would not be possible via the limited computing power and RAM of a consumer gaming system using a typical residential Wi-Fi connection.
But with Gaussian splatting, game designers could achieve photo-realistic visuals that could be updated smoothly in real time, all with a lightweight framework behind the scenes. It’s like opening the hood of a NASCAR vehicle only to find that it is powered by a tiny engine the size of a loaf of bread. How is all of that power possible from only that?
The same capabilities that make it a game-changer for video games make it a powerful tool for other industries, especially because it is readily available in open-source AI tools like OpenSplat and gsplat.
THE BENEFITS OF GAUSSIAN SPLATTING
Photorealistic visualization: One of the most impactful use cases involves using these tools to create photo-realistic 3D visualization of infrastructure. Using a combination of location-based data from photos, GIS systems, CAD drawings, asset management systems, and other sources, organizations can create remarkably detailed digital models of their infrastructure—all without the highly expensive HD scans and modelling that would previously be required to achieve this level of detail.
Real-time rendering: Just as importantly, AI tools that use Gaussian splatting technology can continuously be updated in real time to reflect the latest input, whether that is a new image taken by a work crew, a scan taken by a remote sensor, or another data point. Instant real-time rendering ensures that an organization has a continuously updated digital model of its infrastructure and assets—something that was almost impossible with prior technologies, which typically produced static snapshots of a moment in time.
Mobility: Mobility is also a key advantage. Because these tools can accomplish this real-time visual rendering with a lightweight structure underneath it, workers can put them to use even on devices and in environments that would have been impossible in the past. In the same way that immersive, dynamic gameplay environments can be rendered on an Xbox or PlayStation, workers can interact with ultra-high resolution, real-time digital recreations of critical infrastructure using their smartphone, tablet, or commercially available image capture solutions like Polycam.
HOW ORGANIZATIONS CAN PUT SPLATTING TO WORK
A powerful example is how utilities can utilize Gaussian splatting-powered applications for far more efficient infrastructure maintenance. This is an issue that is growing in importance as more and more utility workers reach retirement age and take a career’s worth of firsthand information about utility assets with them.
Using these AI tools and location intelligence systems (which derive insights from the massive volume of location-based information that organizations can tap into), utilities can create vivid digital models of equipment that may previously have been trapped in the memories of certain workers who were most familiar with the site.
This can drive proactive monitoring of equipment, remote diagnosis and repair of emerging issues, reduced needs for truck rolls, faster response to outages, lower maintenance costs, and the ability to manage more infrastructure with fewer workers. It also excels at change detection, which aids in rapidly identifying issues such as an oil patch, a leak, rust, corrosion, or other equipment issues.
Gaussian splatting tools are also game-changers because they enable organizations with one of the challenges that can slow or undermine AI projects: issues with data availability and data quality. The mathematics in Gaussian splatting is extraordinarily powerful for connecting the dots between available data (even when it is limited and disparate) and filling in those gaps in ways that create an accurate visual model of the real world.
This is a major strength for use cases like the infrastructure management one I discussed above. Visual data and location-based data for an equipment installation or a site may be limited or vary greatly in quality, but it can still successfully be used to create a robust 3D model. Then that model can be improved in real time as newer, better data is collected by crews, sensors, etc.
THE POWER OF SPLATTING
Gaussian splatting is a powerful tool based on three decades of research and refinement. And like many other open-source AI tools, it can be rapidly deployed to solve urgent business problems without the need for a massive AI budget.