Andreas Geiger, leader of the Max Planck Research Group for Autonomous Vision and professor of Computer Science at the University of Tübingen, has received an ERC Starting Grant.
The project funding of 1.47 million euros will run for a period of five years. Starting Grants aim to provide young scientists with greater independence in their research careers.
Geiger will use the grant for his "LEGO 3-D - Learning Generative 3D Scene Models for Training and Validating Intelligent Systems" project. In so doing, the computer scientist aims to develop models that improve the perception of autonomous vehicles and other AI systems, making it easier to monitor their behavior and functioning.
AI systems are based on machine learning. For instance, for an autonomous vehicle to recognize objects in its environment, the corresponding algorithm is trained in the training phase using many sample images. Here, the aim is to make the algorithm work reliably in practice. Until now, the position of objects in the image has had to be determined manually, and this is a cumbersome process. Assessing the behavior of such autonomous systems poses another challenge: often, a static data set is not sufficient because a system’s actions can influence its environment and change it in an unpredictable way.
To overcome these hurdles and make autonomous vision better and more reliable, Geiger wants to develop models that can generate image data automatically and in real time. To this end, the computer scientist first wants to use algorithms to capture the three-dimensional environment and break it down into its elementary components (objects). Based on this disassembling, probabilistic generative models will be trained that are able to reassemble new scenes in a realistic way. The models could be used to generate training data or to test the function of autonomous systems. The results of the research project are not only important for highly automated driving, but could also be used in personal assistance systems, production processes, as well as in entertainment and education.
Andreas Geiger studied computer science and mathematics at the Karlsruhe Institute of Technology (KIT), with research semesters at the Ecole Polytechnique Fédérale in Lausanne (Switzerland) and the Massachusetts Institute of Technology (USA). After completing his doctorate at KIT, he did research in the Perceiving Systems Department att MPI-IS and was appointed Max Planck Group Leader and Visiting Professor at ETH Zurich in 2016. Since March 2018 he has been Professor for "Learning-based Computer Vision" at the University of Tübingen.