Noise2Inverse

This code accompanies

Hendriksen, A. A., Pelt, D. M., & Batenburg, K. J. (2020). Noise2Inverse: self-supervised deep convolutional denoising for linear inverse problems in imaging. CoRR, (), .

Installation

On Linux

Create a conda environment with:

conda env create -f environment.yaml
conda activate noise2inverse
# Install noise2inverse package in environment
pip install -e .

Please be sure to use the exact ASTRA Toolbox version, since newer versions can result in different pixel intensities.

On Windows

There is no windows build of the MSD network available for Windows. We have added an environment file for Windows that does not include the MSD network.

Create a conda environment with:

conda env create -f environment_windows.yaml
conda activate noise2inverse
# Install noise2inverse package in environment
pip install -e .

In the training and evaluation notebooks, make sure to select UNet or DnCNN as the network. The "msd" option is not available. The repository does not contain a trained network weights for UNet and DnCNN. To evaluate the results, you must first run training.

How to use

The notebooks describe how to use the package:

01_generate_projections.ipynb: Generates clean and noisy projections of the foam phantom;
02_reconstruct.ipynb: Contains code for reconstruction;
03_train.ipynb: Trains the network;
04_evaluate.ipynb: Applies the trained network to the noisy reconstructions to obtain a denoised output.
05_metrics.ipynb: Describes metric calculation.

Metrics and results may slightly differ from reported results in paper: the network was retrained using this cleaned up code.

Data