Dr.-Ing. Jan Eric Lenssen receives ECVA PhD Award 2024

Dr.-Ing. Jan Eric Lenssen wrote his dissertation on "Differentiable Algorithms with Data-driven Parameterization in 3D Vision" at the Chair of Computer Graphics at the Department of Computer Science at TU Dortmund University under the supervision of the former chair holder, Prof. Dr. Heinrich Müller.

The dissertation is part of the research field of deep learning (DL) for three-dimensional machine vision. It is based on the observation that the direct transfer of two-dimensional DL methods to three dimensions generally leads to high computing time and memory requirements. It reduces the amount of memory required by using a specific representation of 3D geometry based on a novel "Geometric Graph Neural Network". To reduce the time required, it pursues the concept of hybrid algorithms, which combine knowledge of the problem structure in the form of algorithms with fixed functionality with the representational power of DL models, taking into account the DL requirement of mathematical differentiability, in order to find a favorable trade-off in terms of efficiency, quality and interpretability. For implementation in the field of three-dimensional machine vision, the dissertation presents building blocks of a framework with corresponding GPU implementations. It demonstrates their viability through novel methods for two well-known tasks, for which it achieves better solutions than before. The results of the dissertation were consistently recognized by contributions to leading international conferences in machine vision and machine learning with remarkably high citation frequencies.

The official award was presented at the European Conference on Computer Vision 2024 in Milan (29.09.-04.10.2024). Further information can be found at: https://www.ecva.net