END-TO-END LEARNING OF VARIATIONAL MODELS AND SOLVERS FOR THE RESOLUTION OF INTERPOLATION PROBLEMS - IMT Atlantique Access content directly
Conference Papers Year : 2021

END-TO-END LEARNING OF VARIATIONAL MODELS AND SOLVERS FOR THE RESOLUTION OF INTERPOLATION PROBLEMS

Abstract

Variational models are among the state-of-the-art formulations for the resolution of ill-posed inverse problems. Following recent advances in learning-based variational settings, we investigate the end-to-end learning of variational models, more precisely of the regularization term given some observation model, jointly to the associated solver, so that we can optimize the reconstruction performance. In the proposed end-to-end setting, both the variational cost and the gradient-based solver are stated as neural networks using automatic differentiation for the latter. We consider an application to inverse problems with incomplete datasets (image inpainting and multivariate time series interpolation). We experimentally illustrate that this framework can lead to a significant gain in terms of reconstruction performance, including w.r.t. the direct minimization of the variational formulation derived from the known generative model.
Fichier principal
Vignette du fichier
learning4DVar (22).pdf (234.15 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03139133 , version 1 (26-03-2021)

Identifiers

Cite

Ronan Fablet, Lucas Drumetz, François Rousseau. END-TO-END LEARNING OF VARIATIONAL MODELS AND SOLVERS FOR THE RESOLUTION OF INTERPOLATION PROBLEMS. ICASSP 2021 : IEEE International Conference on Acoustics, Speech and Signal Processing, Jun 2021, Toronto, United States. ⟨10.1109/ICASSP39728.2021.9414629⟩. ⟨hal-03139133⟩
123 View
169 Download

Altmetric

Share

Gmail Facebook X LinkedIn More