Volumetrische Strömungsrekonstruktion mittels differenzierbarer Physik und Selbstüberwachung
Abstract
This dissertation is concerned with reconstructing the turbulent motions of real-world flows. While there are a multitude of classical approaches, gradient-based methods from deep learning allow differentiable simulators to formulate a physically coherent transport that can be optimized end-to-end. With the inclusion of learned self-supervision and multi-scale generator networks, the reconstruction framework can be trained purely from single-view data, without the need for 3D ground truth.
Fields of Study & Subjects
Model Reduction and Neural Networks
Advanced Vision and Imaging
Fluid Dynamics and Turbulent Flows
Publication details
- Journal / source
- mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich)
- Publication date
- 29 January 2026
- Language
- EN
- Document type
- Thesis