In this PGMO lecture I will talk about (non-) convex and non-smooth optimization
methods with applications to variational problems in computer vision. The course
adresses accelerated gradient descent methods and their generalization to
non-convex optimization problems, primal dual methods, and hybrid algorithms
that combine continuous optimization with dynamic programming. I will show how
to solve practical problems that occur in computer vision, such as image
restoration, stereo, optical flow, and medical image reconstruction. In addition
to more theoretical considerations, this lecture will also show how some of the
algorithms can be practically implemented in Python (using jupyter notebooks) .
Participants are therefore invited to bring their laptop with a running version
of Python to participate in the practical tasks. Finally, will also show how to
use machine learning techniques to learn better models from data.
The files (Jupyter notebooks and slides) are available (links at the bottom right of this page).
