As one of the first to join the Third Dimension AI team, Gernot is on the research and engineering team, focusing on dynamic reconstructions for customers.
Can you tell us about your background? What were you working on prior to joining Third Dimension?
I studied computer science at Graz University of Technology and finished my PhD with a focus on computer vision and machine learning. My research revolved around depth and 3D data, from low-level vision to perception, and during this time I also had the opportunity to do an internship at the Max Planck Institute for Intelligent Systems with Andreas Geiger, where we were working towards making 3D computer vision more efficient.
Following my PhD and before joining Third Dimension AI, I was at various companies, including Intel’s Intelligent Systems Lab, Amazon Prime Air, Unity and Luma AI. I worked on different research topics across these companies, ranging from novel view synthesis, optimizing perception tasks, asset generation and reconstruction with NeRFs and Gaussian Splatting.
You were one of the first employees when Third Dimension started. Why did you decide to join the team?
It was a combination of the right timing and also feeling like there was a fit with the team members and founders. When I first met Tolga, I found his personality compelling. He is an easy person to talk to and especially very open. I was also able to meet other existing team members in person before joining and we all just clicked. What Third Dimension is building is closely aligned to my background, so I decided it was the right topic and right time to join a bunch of great people building something new.
When you first joined Third Dimension, the company was in the very early stages. What might you tell a friend who is considering joining an early stage startup?
Sure, in early stage startups, sometimes there can be uncertainty about what the team is building and how to accomplish what you set out to do. But it’s also an opportunity because it empowers you as an early team member to bring a lot of your ideas to the table and shape the direction of the company. I feel like the Third Dimension team has a very open culture, and especially Piotr, our CTO, is always open to feedback and ideas to make what we are building even better.
Unlike bigger companies, your role might not be as specific, so it’s important to be self-motivated and figure out what to spend your time on to make a good impact.
So what are you working on these days at Third Dimension?
In short, I’m focusing on dynamic 3D reconstruction. So, say that we receive data from a customer building a delivery drone. They have a specific camera setup and from their data stream, they want to be able to reconstruct the static world but also have every dynamic object move in that world with realism. What we are able to offer customers is the ability to go into and move around this scene but also simulate the scene across different timestamps. What I’m currently working on is improving the overall reconstruction fidelity and the abilities to extrapolate in space and time.
So how does what we do differ from the “world models” that other companies are working on?
There isn’t a clear definition of “world models” to me, but I’d say what we are building differs in 2 main ways. Many companies are creating what I’d consider “3D” or “static” worlds, whereas we are doing “4D” or “dynamic” reconstruction. 4D refers to the fact that we’ve created this world, and not only does the environment look real, but the dynamic objects such as humans or cars also move through the world in a realistic way over time (time is the fourth dimension). We are also focusing on reconstructing something that’s anchored in reality first, instead of generating things completely from scratch (which can take a lot more time and also does not yield as realistic results, at least at this stage).
What are the most challenging parts of your work?
This is an extremely new space, so there are lots of interesting challenges. For example, every customer has their unique setup - different cameras, different numbers of cameras, different sensor sets they’re using and so on - in the beginning we had to learn how to build a flexible reconstruction pipeline that handles those different set-ups well. We’re now quite good at reconstructing the data that’s provided to us, but the next frontier is optimizing the novel views that extrapolate drastically, for example the actual drive involves our customer’s car driving behind another car, but in simulation they want to perform a take-over or a different action that involves views that were never in the input data.
What are you most excited about working on in 2026?
Now that Christian joined as our Chief Research Officer, I’m excited for him to bring a lot of new ideas and help push the generative reconstruction part of our pipeline to the next level. We’ve also been growing the team quite a bit, so I’m looking forward to making an even bigger impact on more customers together this year.
—
Thank you, Gernot! If you are interested in learning more about the Third Dimension team, visit our Careers page.
