Master’s Thesis, Spring 2027: 3D Perception and Domain-Adaptive models

Do you want your thesis to end up running in production at a sawmill? Taigatech is offering two master’s thesis projects in computer vision and machine learning, starting in January 2027. You’ll work on real industrial data, with real hardware, alongside the team that built it. Taigatech makes sustainable wood production more competitive. Our AI and computer vision systems help sawmills and pulp mills use more of their raw material and keep their lines running. Founded in 2020 out of Chalmers, we are around fourteen people in Gothenburg, with Fredrik Kahl, professor of computer vision at Chalmers, as one of our founders. Our systems run in production in Sweden and abroad. Project 1: 3D perception – adding depth to our vision systems Our perception models today see the world through cameras. In this project you will find out what depth adds. You’ll help us choose and set up 3D sensor hardware, collect and annotate your own 3D data, and build a 3D perception model for the same task our camera-based models solve. Then you compare the two approaches on the same scenes. If the results look promising, you can go on to explore fusing 3D and image data in a single model. How far you get is shaped by what you find. You will join at the very start: the hardware isn’t bought yet, so you’ll have a real say in what we buy, how it is mounted and how the data is collected. Your results will directly inform which sensing technology we invest in next. Good fit if you like: deep learning on 3D data (3D detection and segmentation), sensors and data collection, careful experimental comparison. Project 2: Domain-adaptive models – perception that works at the next site on day one Our cameras are fixed in place, which means every site has its own static background, lighting and seasons. Models can pick up on that background instead of the objects we care about. The goal of this project is to make our models stay accurate when we move to a new site, a new camera, or a new season, with as little relabelling as possible. You will implement and evaluate modern domain adaptation techniques on our detection and segmentation models. Possible directions include adversarial approaches such as gradient reversal layers and domain-adaptive detectors, self-training, cross-domain data mixing, and conditioning a model on a handful of example images from a new view. You will have plenty of room to experiment and choose your own direction. Good fit if you like: model training and experimentation, reading and reproducing recent papers, robustness and generalisation. Have your own idea? The two projects above are starting points, not fixed assignments. If you want to adjust one to fit your interests, or you have an idea of your own in industrial computer vision, perception or machine learning, we want to hear it. Tell us in your cover letter what you’d like to explore and why. We’ll tell you honestly whether it fits, and we’re happy to shape it together. What you get Real data and real impact. Authentic industrial data from several sites, and a path for your results into a live production system. Hardware you help choose. Direct supervision. Day-to-day support from our ML and computer vision engineers, and academic input through our Chalmers connection. Freedom to shape the work. You’ll have a real say in the direction, from day one. A foot in the door. We are growing, and a thesis is the best way to find out whether we are a good match. We are always interested in talking about a job afterwards. Who we are looking for You are in the final year of a master’s programme in engineering physics, mechatronics, electrical engineering, engineering mathematics, computer science, data science, machine learning or a related field You have a strong foundation in mathematics and machine learning, and have trained neural networks hands-on (PyTorch or similar) You are comfortable programming in Python You learn quickly, take ownership, and can work independently on a hard technical problem Experience with computer vision is a plus; for project 1, experience with 3D data or sensors is a plus You are available to start in January 2027 How to apply Send your CV and a short cover letter telling us which project interests you, or what you’d like to work on instead, and why. We interview on a rolling basis, so don’t wait for the last day.