305. Multilevel Plug-and-Play Image Restoration
Invited abstract in session WB-4: Optimization and learning for estimation problems, stream Optimization for machine learning.
Wednesday, 10:30-12:30Room: B100/5013
Authors (first author is the speaker)
| 1. | Nils Laurent
|
| 2. | Julian Tachella
|
| Physics laboratory, CNRS & ENS Lyon | |
| 3. | Elisa Riccietti
|
| LIP, ENS Lyon | |
| 4. | Nelly Pustelnik
|
| Univ Lyon, Ens de Lyon, Univ Lyon 1, CNRS, Laboratoire de Physique, Lyon |
Abstract
Plug-and-play (PnP) image reconstruction methods leverage pretrained deep neural network denoisers as image priors to solve general inverse problems, and can obtain a competitive performance without having to train a network on a specific problem. Despite their flexibility, PnP methods often require several iterations to converge and their performance can be highly sensitive to the choice of the initialization and of the hyperparameters. In this talk, we propose a new multilevel PnP framework that accelerates the convergence by combining iterations at different scales, i.e. involving different resolutions of the image, and improves the robustness to initialization and hyperparameters setting using a coarse-to-fine strategy. In a series of experiments, including image inpainting and demosaicing, we show that the proposed multilevel PnP method outperforms other PnP methods in both speed and reconstruction performance.
Keywords
- AI based optimization methods
- Multi-level optimization
- Complexity and efficiency of algorithms
Status: accepted
Back to the list of papers