Skip to Main content Skip to Navigation
Conference papers

A Flexible Framework for Designing Trainable Priors with Adaptive Smoothing and Game Encoding

Bruno Lecouat 1, 2 Jean Ponce 1 Julien Mairal 2
1 WILLOW - Models of visual object recognition and scene understanding
Inria de Paris, DI-ENS - Département d'informatique de l'École normale supérieure
2 Thoth - Apprentissage de modèles à partir de données massives
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann
Abstract : We introduce a general framework for designing and training neural network layers whose forward passes can be interpreted as solving non-smooth convex optimization problems, and whose architectures are derived from an optimization algorithm. We focus on convex games, solved by local agents represented by the nodes of a graph and interacting through regularization functions. This approach is appealing for solving imaging problems, as it allows the use of classical image priors within deep models that are trainable end to end. The priors used in this presentation include variants of total variation, Laplacian regularization, bilateral filtering, sparse coding on learned dictionaries, and non-local self similarities. Our models are fully interpretable as well as parameter and data efficient. Our experiments demonstrate their effectiveness on a large diversity of tasks ranging from image denoising and compressed sensing for fMRI to dense stereo matching.
Complete list of metadatas

Cited literature [87 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02881924
Contributor : Bruno Lecouat <>
Submitted on : Wednesday, October 28, 2020 - 11:07:53 AM
Last modification on : Tuesday, November 24, 2020 - 4:00:09 PM

File

main.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02881924, version 2

Collections

Citation

Bruno Lecouat, Jean Ponce, Julien Mairal. A Flexible Framework for Designing Trainable Priors with Adaptive Smoothing and Game Encoding. Conference on Neural Information Processing Systems (NeurIPS), Oct 2020, Vancouver, France. ⟨hal-02881924v2⟩

Share

Metrics

Record views

34

Files downloads

36