eugenepalovcak / restore

A CNN for denoising cryo-EM images

UCSF restore

restore is a program for denoising cryogenic electron microscopy images with a convolutional neural network.

Requirements

restore requires a Linux system with an NVIDIA GPU. MotionCor2 is recommended for generating the training data.

Installation for Linux

Working on a simpler installation. For the time being:

1. Make sure the NVIDIA driver is up-to-date. (Must be >410)
2. Install Miniconda3 (or Anaconda) if necessary.
3. Make sure git is installed.
4. Download the package and navigate into the directory:

git clone https://github.com/eugenepalovcak/restore
cd restore

5. Create a conda python environment that contains the required dependencies. This step may take a few minutes.

conda env create -f restore.yml
conda activate restore

6. Now install Daniel Asarnow's pyem library for cryo-EM data processing

git clone https://github.com/asarnow/pyem.git
pip install ./pyem

7. Now, install restore:

pip install .

and append the restore directory to the PATH variable in your .bashrc file:

restore_directory=`pwd`
echo 'export PATH=DIR:$PATH' | sed "s#DIR#${restore_directory}#" >> ~/.bashrc

Make sure you use '>>' and not '>' or you'll overwrite your bashrc instead appending to it! Source the .bashrc and you should be good to go!

source ~/.bashrc

8. Whenever you want to run restore, you simply need to activate the restore python environment:

conda activate restore

and you should be able to run the executables train.py and denoise.py When you're done using restore, you can deactivate the environment:

conda deactivate