RD-DIP: Rician Denoising - Deep Image Prior

by HR Iglesias-Goldaracena, I. Ramírez & E. Schiavi.

This repository contains the dependencies, code and datasets for harnessing RD-DIP, a major enhancement for the unsupervised learning Deep Image Prior (DIP) that allows it to perform Rician denoising.

Try it without installing anything—click here to open the project in Google Colab!

Graphical abstract

Visualization of the (iterative) denoising process

Ground-truth	Noisy ($\sigma=0.15$)
Baseline DIP	RD-DIP

Getting started

Clone the repository and initialize the modules:

git clone [email protected]:heqro/rd-dip.git
cd rd-dip
git submodule init
git submodule update --recursive --remote

Now, create a virtual environment, activate it and then and install the dependencies:

python -m venv .venv
source .venv/bin/activate 
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install piqa
pip install scikit-image

Note that the previous codeblock installs torch with the CUDA 11.8 version, as it is the version this software package was developed with.