Generative Models as Regularizers for Inverse Problems in Imaging

This thesis encompasses largely the two articles:

• "Stable Deep MRI Reconstruction using Generative Priors"

  • Code
  • Article

• "Product of Gaussian Mixture Diffusion Models"

  • Code
  • Article

The source code needs to be compiled with lualatex and uses the tikz externalization library for parallel processing and caching of figures. Thus, compilation looks something like this:

# first latexmk run, populates `main.makefile` with code to produce figures
latexmk -shell-escape -lualatex main.tex 
# make figures
make -j -f main.makefile
# second latexmk run
latexmk -shell-escape -lualatex main.tex