Évènement

Ségolène Martin
TU Berlin
https://segolenemartin.github.io

Date(s) : 11/05/2026   iCal
14h00 - 15h00

We begin this talk with an introduction to generative modeling for image synthesis. Starting from discrete normalizing flows, we move to continuous normalizing flows and finally to flow matching. We clarify these approaches, highlighting their specificities, limitations, and provide simple implementation insights. We also discuss their connections with diffusion models.

In the second part, we explore how these models can be used to solve linear inverse problems in imaging, such as image inpainting, deblurring, and super-resolution. We emphasize the connections between flow matching and denoising, and show how these links can be leveraged within the Plug-and-Play (PnP) framework to address inverse problems.

Emplacement
Saint-Charles - FRUMAM (2ème étage)

Catégories

• Séminaire