With newsletter #7 we are entering the home stretch of the KI Familie projects. It is probably the most interesting phase for the AI community as we are presenting more and more results of this unique cluster. One of the four projects - KI Absicherung - was already completed 2022, the three sister projects KI Delta Learning, KI Wissen, and KI Delta Tooling keep on working tirelessly, achieving milestone by milestone, and are about to prepare their final events in 2023.
In this edition, we look back to the second Open Project Day of KI Wissen, which took place at Fraunhofer in Sankt Augustin in January, focusing on the progress made in training efficiency by including already available knowledge and context data; and forth to KI Delta Learning – the next project to reach the finish line. The presentation of KI Delta Learning’s results is scheduled for March 9 (public day) and 10 (expert / family day). If you cannot make it to the Mercedes-Benz premises in Stuttgart-Vaihingen in March, join online (dial-in information below) or stay tuned for the next newsletter! Last but not least, KI Data Tooling reveals interesting project findings in the field of advanced AI training strategies… So please have a closer look and enjoy reading!
The KI Familie Editorial Team
KI Wissen: Second Turn of Open Project Days in Sankt Augustin
On January 17, 2023, the partners of the KI Familie projects (KI Wissen, KI Delta Learning, KI Data Tooling) have met to exchange information on efficient and knowledge-based AI systems. More than 90 attendees from industry, SMC and science took the opportunity to get an update on the overall project progress as well as to discuss details with partners directly and in person.
We are right in the middle of the preparations for the KI Delta Learning’s Final Event.
The Mercedes-Benz Campus in Stuttgart-Vaihingen will be the venue where dozens of project members from all over Germany will come together on March 9 and 10, 2023. The event will focus on presenting the results of the BMWK funded research project in high-class talks, posters and presentations. The first day will be public and gives insights into the project’s key achievements on the path towards open world generalization. The second day is designed for members of the KI Familie to discuss the results in deep dives and to push for “transfer learning” within the cluster of fellow projects.
One important development step of the KI Delta Learning project was completed successfully by the end of 2022: The recording of real driving data in Italy complemented the national data recording.
A test vehicle equipped with numerous sensors including cameras and lidar drove on pre-defined routes on public streets in Italy during November 2022. Prior to this driving experience and real data recording, many planning steps were taken.
Picture: Deltas: New domains, complex situations require step-by-step extensions
Creating a Data Kit for Advanced AI Training Strategies
Imagine having access to a digital replica of a real-world traffic scene that looks and behaves like the real thing. This is no longer a fantasy, as recent advances in computer graphics and simulation technologies have made it possible to create high-resolution digital twins of real traffic scenes. These digital twins are crucial for the development of next-level data-driven automated driving functions. The KI Data Tooling project is at the forefront of this development, creating a pioneering data kit comprising a high-quality dataset and a collection of realistic digital twins. What is s even more exciting is that the data kit includes both real and synthetic versions of the same scene, captured across all relevant sensor domains. With this data kit, the project aims to develop efficient and effective AI training strategies for the development of advanced automated driving functions.
Intelligent Vehicle Symposium 04.-07. June Anchorage, Alaska Find out more
CVPR 2023 Conference on Computer Vision and Pattern Recognition 18.-20. June Vancouver, Canada Find out more
ICPRS 2023 International Conference on Pattern Recognition Systems 04.-07. July Guayaquil, Ecuador Find out more
ARTS23 Automated Road Transportation Symposium 09.-13. July San Francisco, USA Find out more
ICECCME 2023 International Conference on Electrical, Computer, Communications and Mechatronics Engineering 19.-20. July Tenerife, Canary Islands, Spain Find out more
IAA Mobility 2023 05.-10. September Munich, Germany Find out more
Academic Corner - KI Familie Publications
Domain Adaptation and Generalization: A Low-Complexity Approach Joshua Niemeijer, Jörg P. Schäfer Read more
Heatmap-based Out-of-Distribution Detection Julia Hornauer Read more
iOn the calibration of underrepresented classes in LiDAR-based semantic segmentation Marielle Dreissig, Florian Piewak, Joschka Boedecker Read more
SynPeDS – A Synthetic Dataset for Pedestrian; Detection in Urban Traffic Scenes Thomas Stauner, Frederik Blank, Michael Fürst, Johannes Günther, Korbinian Hagn, Philipp Heidenreich, Markus Huber, Bastian Knerr, Thomas Schuli, Karl Leiss Read more
Towards Runtime Monitoring of Complex System Requirements for Autonomous Driving Functions Dominik Grundt, Eike Möhlmann, Anna Köhne, Ishan Saxena, Ralf Stemmer, Bernd Westphal Read more
Suppress with a Patch: Revisiting Universal Adversarial Patch Attacks against Object Detection Svetlana Pavlitskaya, Jonas Hendl, Sebastian Kleim, Leopold Johann Müller, Fabian Wylczoch, J.Marius Zöllner Read more
A Benchmark for Unsupervised Anomaly Detection in Multi-Agent Trajectories Julian Wiederer, Julian Schmidt, Ulrich Kreßel, Klaus Dietmayer, Vasileios Belagiannis Read more
Adversarial Vulnerability of Temporal Feature Networks for Object Detection Svetlana Pavlitskaya, Nikolai Polley, Michael Weber, J.Marius Zöllner Read more
An Unsupervised Domain Adaptive Approach for Multimodal 2D Object Detection in Adverse Weather Conditions George Eskandar, Robert A. Marsden, Pavithran Pandiyan, Mario Döbler, Karim Guirguis, Bin Yang Read more
Validation of pedestrian detectors by classification of visual impairing factors Korbinian Hagn, Oliver Grau Read more
Two Video Data Sets for Tracking and Retrieval of Out of Distribution Objects Kira Maag, Robin Chan, Svenja Uhlemeyer, Kamil Kowol, Hanno Gottschalk Read more
One Ontology to Rule Them All: Corner Case Scenarios for Autonomous Driving Daniel Bogdoll Read more
A-Eye: Driving with the Eyes of AI for Corner Case Generation Kamil Kowol, Prof. Stefan Bracke, Prof. Hanno Gottschalk Read more
SynPeDS – A Synthetic Dataset for Pedestrian; Detection in Urban Traffic Scenes Thomas Stauner, Frederik Blank, Johannes Günther, Michael Fürst, Korbinian Hagn, Philipp Heidenreich, Markus Huber, Bastian Knerr, Thomas Schulik, Karl Leiss Read more
Infusing Contextual Knowledge Graphs for Visual Object Recognition Sebastian Monka, Lavdim Halilaj Read more
Semi-supervised domain adaptation with CycleGAN guided by a downstream task loss Annika Mütze, Matthias Rottmann, Hanno Gottschalk Read more
Automated Detection of Label Errors in Semantic Segmentation Datasets via Deep Learning and Uncertainty Quantification Matthias Rottmann, Marco Reese Read more
Improving Replay-Based Continual Semantic Segmentation with Smart Data Selection Tobias Kalb, Björn Mauthe, Jürgen Beyerer Read more
Causes of Catastrophic Forgetting in Class-Incremental Semantic Segmentation Tobias Kalb, Jürgen Beyerer Read more