In this team spotlight, we get to know Jeroen Claessens as he shares his journey into computer science and data labeling tooling.
Q: “On land, at sea, and in the air”, would that be a good catchphrase for your career?
I never thought of it that way, but that indeed applies to me. After my studies, I ventured to Finland. There, I interned with Norsepower. Even today, I am still actively contributing to the project.
Norsepower, founded in late 2012, is dedicated to reducing the environmental impact of shipping. Their flagship innovation, the Norsepower Rotor Sail, combines traditional sailing principles with innovative technology.
Upon returning to Belgium, I joined Be-Mobile, the company behind applications like 4411 (mobility and parking) and Flitsmeister (real-time traffic information and navigation). I joined as a frontend developer with a project on visualizing traffic data & mobility indicators in charts and on maps. For example, you then could see how many pedestrians, bicycles, and public transport are active within different parts of the city at different times of the day.
A year later, I joined SkyeBase, an Antwerp startup specializing in drone inspections. At SkyeBase, one of the projects I worked on was building an in-house data annotation tool. It was similar to Segments.ai, but a light version.
With other priorities in features, we left the labeling tool at a very basic version. We noticed that maintaining such a tool with all the features you would expect is very time-intensive. And it takes away time to develop the core of the platform. I am currently writing down some of the learnings from that project. I’ll add a link here soon.
That project matched my passion for visualizing data. So when I saw an opening at Segments.ai pass by on LinkedIn, I knew I had to go for it. The context of a 3D data labeling platform was especially appealing to me, given my previous experience, where I realized the complexities involved in processing and connecting 3D data. Working with visual elements, especially in computer vision, is inherently more intriguing and satisfying for me compared to dealing with mere text or formulas.
Q: You mentioned the development of an internal labeling platform. Can you share the factors influencing your decision to create an internal platform instead of using an existing solution?
Our requirements centered around two core functionalities: data labeling and visualization.
We had an AI engineer responsible for training models. For that need, we decided upon an external labeling platform.
For our data visualization needs, we mainly needed bounding boxes and vectors to assist in tasks like defect detection. For example, corrosion detection on steel cranes or cracks in concrete bridges. While searching for suitable labeling tools, we talked with Segments.ai. The choice was quite balanced between building our own solution or opting for licensing.
Given that our requirements seemed straightforward at first, we ultimately chose to build the tool in-house, as it appeared more feasible and aligned with our budget constraints.
However, there were notable differences in performance between our in-house solutions and those available externally. For instance, we developed a segmentation feature similar to the Superpixel. But in comparison, our in-house version had several areas where there was room for performance improvements, such as taking seconds to overlay an image. In contrast, the Segments.ai version is instant, scrollable, and better adapted to the scene. But that is the trade-off between in-house development and external solutions, particularly in terms of performance and efficiency versus costs.
Q: What was the first project you worked on at Segments.ai for the labeling platform?
The first project for the data labeling was to improve the polygons tool in the image labeling interface. You could draw polygons, but you couldn’t manipulate them easily afterward. For example, moving them or adding and removing points. At the same time, it was crucial to ensure that the user experience remained consistent with our point cloud interface, which already has this feature.
Next, I integrated a ruler tool for image and point cloud interfaces. In 2D, the ruler measures the pixel distance between objects. However, in 3D, it provides measurements in actual units like centimeters or meters. This is particularly useful in scenarios like tracking a car’s movement in a sequence, allowing us to quantify how much it has moved precisely.
Q: On land, at sea, and in the air. What recent trends in technology do you find fascinating?
The most captivating trend I’ve observed recently is the evolution of self-driving cars, robots, and drones. These technologies have been around for a while, but their progression is remarkable. Initially, we saw partial automation, such as assisted steering and acceleration. However, the shift we’re witnessing now is towards fully automated systems capable of completely replacing human intervention.
This trend is becoming increasingly significant in various sectors. For example, autonomous vehicles like Cruise and Waymo taxis, operating round-the-clock, are now common in some cities. Then there’s Carken, a remarkable service that handles over 400 daily deliveries at just one university. And Zipline, which is revolutionizing medical delivery in several countries.
Q: Have you noticed any impact of AI in your day-to-day work as a developer?
Interestingly, I don’t consciously use AI in my daily work, but I actually do. I regularly use GitHub Copilot, an extension in Visual Studio Code. It significantly aids in development tasks. While I don’t always actively think about using it, I immediately notice something missing when the extension is not installed or disabled when coding. It functions like an advanced autocomplete, providing context-sensitive suggestions. For instance, it can adjust lines of code by understanding the context, offering subtly different yet relevant variations.
I’ve been using GitHub Copilot for about two years, and it has seamlessly integrated into my workflow to the point where I often use it without realizing it. So, yes, AI does impact my day-to-day work through tools like GitHub Copilot.
Q: How do you envision your ideal home office setup as a fully remote worker?
My current home office setup, in terms of screens and accessories, is complete. I have three large screens, and my laptop is in the background. Recently, I’ve started using a GoPro as a webcam, which is still a work in progress, especially adjusting all the settings and figuring out the proper zoom levels.
However, there’s room for improvement in aesthetics and comfort. The space is bare, with just my desk and screens against plain white walls. I’m considering adding some decorations to make it more inviting and cozy. Incorporating different lighting options and wall decorations, like a map, would be ideal. I love traveling, so having a map on the wall would be a nice touch, reflecting my passion for exploring new places.
These trips are my way of detoxing from constant screen time and the demands of being connected for work. I want to explore more of the world beyond just America and Europe.
Q: What are your other hobbies or activities outside of work besides traveling?
I’ve been an avid basketball player for almost 20 years until last year. Balancing basketball with work became challenging since I had to train twice a week and attend matches on fixed days. Seeking more flexibility, I decided to switch to karting, which I got into through a connection in motorsport. Karting offers the thrill and excitement I enjoy but with a schedule that’s easier to manage alongside my professional commitments. It’s really one of the only things where I completely can shut of my mind and not think about anything else apart from the flowing race track and beating track times.
One of my major passions is cars. I own an electric car, but my heart truly lies with traditional combustion engine vehicles. If I were born earlier, I would have owned one of those classic gasoline cars from the 80s or 90s.
My dream car is the Audi RS6. It’s a station wagon, which might seem different from the typical enthusiast’s choice, but it’s a perfect blend of utility and performance. It’s large and practical, allowing you to carry everything you need, yet it doesn’t compromise on speed. It’s not like a Lamborghini, where you sacrifice practicality for performance. With the RS6, you get the best of both worlds. Although, I’m very happy to drive my i4, which at the moment is my dream car if it has to be electric.