Column-Action Methods in Image Reconstruction
Reconstructing real fluid phenomena based on sparse input images is a very challenging topic. In this thesis, we focus on reconstructing the 3D density volume of a rising smoke plume and neglect the reconstruction of the 3D motion. Computed Tomography is the standard approach for reconstructing 3D volumes based on 2D input images. A system of linear equations is constructed and solved. In order to capture the fine swirls of real smoke, we need a fine resolution of our discretized 3D volume. The higher the resolution is, the more memory and runtime are required for solving the tomography equations, which is limiting the resolution in practice to an unsatisfying level.
The goal of this thesis is to solve the tomography equations with column-action methods. They converge to least squares solutions and save computational work by skipping small updates. A more efficient solver (memory- and runtime-wise) allows us to reconstruct 3D volumes on finer grids.
- Interest in numeric solvers
- C++, fundamental knowledge in python, advanced math skills
- Advantageous: experiences with mantaflow
- Please provide your CV and transcript of records
- Ihrke, I. and M. Magnor (2004). Image-based tomographic reconstruction of flames. Proceedings of the 2004 ACM SIGGRAPH/Eurographics symposium on Computer animation, Eurographics Assoc.
- Gregson, James, et al. "Stochastic tomography and its applications in 3D imaging of mixing fluids." ACM Trans. Graph. 31.4 (2012): 52-1.