Divergence-Free Neural Networks with Application to Image Denoising

Sébastien Herbreteau and Etienne Meunier

News

• 🔥 Accepted at ICLR'26

Requirements

Here is the list of libraries you need to install to execute the code:

• Python 3.8
• Pytorch 2.8
• DeepInverse 0.3.4

To install in an environment using conda:

conda env create -f environment.yml
conda activate censure
pip install deepinv

Demo

We provide a Python Jupyter Notebook with example code to reproduce the experiments of the paper: demo.ipynb.

CENSURE model

Our proposed parameterization of divergence-free neural networks is implemented in Pytorch in the file censure.py.

The pre-trained weights are available at: https://github.com/sherbret/divergence_free_nn/releases/tag/v1.0 (to be downloaded and placed in the folder pretrained_weights).

Acknowledgements

The authors acknowledge the support of the Research Group IASIS of the CNRS Informatics (Project DIVIN). Etienne Meunier was partially supported by the Research Programme "PPR Ocean & Climate" through a postdoctoral scholarship. This work was granted access to the HPC resources of IDRIS under the allocation 2025-AD011015932R1 made by GENCI. We thank Cédric Herzet for fruitful discussions.

Citation

@inproceedings{
herbreteau2026divergencefree,
title={Divergence-Free Neural Networks with Application to Image Denoising},
author={S{\'e}bastien Herbreteau and Etienne Meunier},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=a5lL1ygtkG}
}