Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeepInverse: A Python package for solving imaging inverse problems with deep learning

Julián Tachella, Matthieu Terris, Samuel Hurault, Andrew Wang, Dongdong Chen, Minh-Hai Nguyen, Maxime Song, Thomas Davies, Leo Davy, Jonathan Dong, Paul Escande, Johannes Hertrich, Zhiyuan Hu, Tobías I. Liaudat, Nils Laurent, Brett Levac, Mathurin Massias, Thomas Moreau, Thibaut Modrzyk, Brayan Monroy, Sebastian Neumayer, Jérémy Scanvic, Florian Sarron, Victor Sechaud, Georg Schramm, Romain Vo, Pierre Weiss

TL;DR

DeepInverse addresses the fragmentation of learning-based imaging inverse problem methods by providing an open-source, PyTorch-based library that unifies forward operators, variational formulations, and a broad family of solvers. It covers optimization-based, sampling-based, and non-iterative approaches, complemented by a training framework, standardized losses, datasets, and evaluation metrics. The design emphasizes differentiable, matrix-free operators, multiple noise models, and reproducibility through rigorous coding practices and comprehensive documentation. By enabling standardized operators, losses, datasets, and evaluators, DeepInverse aims to accelerate cross-domain research and practical adoption of deep learning methods for image reconstruction.

Abstract

DeepInverse is an open-source PyTorch-based library for solving imaging inverse problems. The library covers all crucial steps in image reconstruction from the efficient implementation of forward operators (e.g., optics, MRI, tomography), to the definition and resolution of variational problems and the design and training of advanced neural network architectures. In this paper, we describe the main functionality of the library and discuss the main design choices.

DeepInverse: A Python package for solving imaging inverse problems with deep learning

TL;DR

DeepInverse addresses the fragmentation of learning-based imaging inverse problem methods by providing an open-source, PyTorch-based library that unifies forward operators, variational formulations, and a broad family of solvers. It covers optimization-based, sampling-based, and non-iterative approaches, complemented by a training framework, standardized losses, datasets, and evaluation metrics. The design emphasizes differentiable, matrix-free operators, multiple noise models, and reproducibility through rigorous coding practices and comprehensive documentation. By enabling standardized operators, losses, datasets, and evaluators, DeepInverse aims to accelerate cross-domain research and practical adoption of deep learning methods for image reconstruction.

Abstract

DeepInverse is an open-source PyTorch-based library for solving imaging inverse problems. The library covers all crucial steps in image reconstruction from the efficient implementation of forward operators (e.g., optics, MRI, tomography), to the definition and resolution of variational problems and the design and training of advanced neural network architectures. In this paper, we describe the main functionality of the library and discuss the main design choices.

Paper Structure

This paper contains 29 sections, 7 equations, 1 figure, 2 tables.

Table of Contents

  1. Statement of Need
  2. Inverse Problems
  3. Operator Parameterization
  4. Reconstruction Methods
  5. Optimization-Based Methods
  6. Explicit Priors
  7. Denoising Plug-and-Play Priors
  8. Optimization Algorithms
  9. Unfolded Networks and Deep Equilibrium Models
  10. Sampling-Based Methods
  11. Diffusion Models
  12. Langevin-Type Algorithms
  13. Non-Iterative Methods
  14. Artifact Removal
  15. Unconditional Generative Networks
  16. ...and 14 more sections

Figures (1)

  • Figure 1: Schematic of the main modules of the library.