BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Solving image inverse problem with deep generative models
DTSTART:20260630T091500Z
DTEND:20260630T101500Z
DTSTAMP:20260810T223100Z
UID:indico-event-14488@indico.math.cnrs.fr
DESCRIPTION:Speakers: Jean Prost (IRIT)\n\nImage inverse problems involve 
 recovering a clean image from a degraded observation. Such problems arise 
 in numerous applications\, including photography\, medical imaging\, and a
 strophysics. These problems are typically ill-posed because part of the im
 age information is lost or corrupted during the acquisition process.Under 
 a Bayesian framework\, the posterior distribution of the solution can be m
 odeled by combining information from the degraded observation with prior k
 nowledge about the solution. Deep generative models provide powerful prior
 s\, but working with the resulting posterior distribution remains challeng
 ing due to the complexity of these models. I will present an optimization
  algorithm for computing an (augmented) maximum a posteriori (MAP) estimat
 or when the prior is induced by a hierarchical variational autoencoder (HV
 AE). The proposed method leverages the HVAE encoder to avoid backpropagati
 on through the generative model\, thereby reducing the computational cost 
 of inference.I will then introduce a posterior sampling algorithm based on
  a text-to-image latent consistency model (LCM). By exploiting the fast ge
 neration capabilities of LCMs\, the method produces high-quality posterior
  samples in only a few iterations (approximately eight). The text prompt c
 an either be specified in advance to guide the reconstruction or automatic
 ally recovered through a prompt-optimization procedure that seeks the most
  likely textual description of a degraded observation. The proposed approa
 ch outperforms competing methods based on other text-to-image generative m
 odels while being significantly faster and more memory-efficient.\n\nhttps
 ://indico.math.cnrs.fr/event/14488/
LOCATION:Salle K. Johnson (1R3\, 1er étage)
URL:https://indico.math.cnrs.fr/event/14488/
END:VEVENT
END:VCALENDAR