During the past few years, my research has focused on two main areas: optimal transport and video processing. My HDR details these two aspects.

Optimal transport is a trending framework for manipulating probability distributions, histograms, or more generally, functions. It consists in seeing a function as a pile of sand moving in space at a minimum cost. This allows, for instance, to define a way to interpolate between two (or more) probability distributions, or to define a meaningful distance between histograms. This theory has seen many applications from computer graphics to deep learning.
Here, my work has first focused on building efficient algorithms to solve optimal transport problems. Then, I defined and solved inverse problems making use of the optimal transport geometry, such as computing barycentric coordinates or performing dictionary learning for histograms.

Video processing has seen a large growth due to the wide availability of consumer cameras, and user-friendly apps featuring powerful video processing capabilities such as SnapChat, Instagram or Tik Tok. However, re-adapting the long history of image processing algorithms to make them work on videos is a daunting task. Indeed, trivially applying image processing filters to all frames of a video most often results in temporal artifacts such as flickering. My work has focused on bringing common image processing algorithms to the realm of videos, in a temporally consistent fashion. I first worked on bringing specific filters to videos (color grading and intrinsic images), and ultimately found a way to bring many image processing filters to videos without even knowing their formulation.

My HDR presentation will describe my most representative contributions.