André Kelm

• Deep learning (DL) for industrial innovation [IGD], with cross-disciplinary interests in physics, audio, medical and biological applications
• Domain adaptation and use of synthetic data for real-world
• Efficient, adaptive inference, as well as interpretability for DL [paper]
• Bottom-up and top-down attention [arxiv]
• High-quality segmentation and object shape extraction using Walk the Lines [WtL-arxiv], [video]
• Edge and object contour detection [RCN-arxiv]

2024

• André Peter Kelm, Niels Hannemann, Bruno Heberle, Lucas Schmidt, Tim Rolff, Christian Wilms, Ehsan Yaghoubi, Simone Frintrop: Dynamic Inference and Top-down Attention in a Hierarchical Classification Network, in Proceedings of the 27th International Conference on Pattern Recognition (ICPR 2024), Lecture Notes in Computer Science, vol. 15328, A. Antonacopoulos, S. Chaudhuri, R. Chellappa, C.-L. Liu, S. Bhattacharya, U. Pal, Eds., Springer, Kolkata, India, December 1–5, 2024,[arxiv]
• André Peter Kelm, Niels Hannemann, Bruno Heberle, Lucas Schmidt, Tim Rolff, Christian Wilms, Ehsan Yaghoubi, Simone Frintrop: Select High-Level Features: Efficient Experts from a Hierarchical Classification Network, 5th Workshop on practical ML for limited/low resource settings (PML4LRS) at ICLR 2024,[arxiv], [PDF]
• Emre Gülsoylu, Paul Koch, Mert Yildiz, Manfred Constapel, André Peter Kelm: Image and AIS Data Fusion Technique for Maritime Computer Vision Applications, 2nd Workshop on Maritime Computer Vision at Winter Conference on Applications of Computer Vision (WACV), 2024,  [PDF], [arxiv], [video]

2023

• Ehsan Yaghoubi, André Kelm, Timo Gerkmann, Simone Frintrop: Acoustic and Visual Knowledge Distillation for Contrastive Audio-Visual Localization, International Conference on Multimodal Interaction, ICMI 2023, [PDF]
• André Peter Kelm, Niels Hannemann, Bruno Heberle, Lucas Schmidt, Tim Rolff, Christian Wilms, Ehsan Yaghoubi, Simone Frintrop: High-Level Parallelism and Nested Features for Dynamic Inference Cost and Top-Down Attention, arXiv:2308.05128, 2023, [arXiv], [PDF]