Projects for Bachelor and Master theses, as well as Guided Research, are always available!
Please contact any of our team members via email to inquire about a potential project.
In general, most projects focus on our research interests and open-source software. Some of our current project ideas are outlined below. Please do not hesitate to contact us, so we can match your interest and requirements to our projects.
Anisotropic X-ray Dark-field Tomography (AXDT) is a novel imaging modality introduced by our group, relating to X-ray small-angle scattering. Potential projects arise in the context of the very ill-posed problem of reconstructing spherical scattering functions in 3D from 2D dark-field projection data, for example:
- Development of statistical reconstruction methods for AXDT: modeling the dark-field signal statistics and developing suitable iterative algorithms for solving the associated inverse problem.
- Accelerating inverse problem solving for AXDT: applying techniques such as preconditioning, approximate gradients or momentum to iterative algorithms, and applying them to real experimental data.
- Stabilizing the inverse problem of AXDT: adapting and implementing spherical function regularization techniques to apply prior knowledge, with validation on real experimental data.
Light field microscopy (LFM) is a fast scan-less volumetric imaging technique that can be applied to the imaging of distributed neuronal activity in fluorescent reporter fish. Potential projects arise in the context of efficiently solving the inverse problem of reconstructing a time-series of 3D volumes from 2D light field images, for example:
- Light field microscopy forward modeling: efficient computation of the light field point spread function for arbitrary microscope camera geometries.
- Accelerating inverse problem solving for LFM: applying techniques such as preconditioning, approximate gradients or momentum to iterative algorithms, and applying them to real experimental data.
- Learning depth from light field: supplying additional knowledge to the inversion process by learning depth from the raw light field images using deep learning techniques.
elsa is a modern, elegant C++ library intended for use in tomographic reconstruction. The open-source library is used in our research projects, and thus constantly under development. Potential projects are:
- Implementation of cudnn-based layers to facilitate using deep learning techniques in elsa.
- High performance computing for discretization of the X-ray transform: efficient GPU-based implementation of advanced discretization techniques of the X-ray transform, potentially involving the ray tracing cores of modern GPU accelerators.
- Implementation of multi-resolution approaches: using techniques like octrees and algebraic multi-grid to enable efficient and flexible computations at varying levels of resolution.
- Efficient rasterization of arbitrarily rotated ellipsoids: enable fast and flexible phantom generation for rapid prototyping and validation.