A small example on how to use CSBDeep to train a network with the Noise2Noise training scheme.
- Install miniconda
- Make sure your GPU supports CUDA 10.
- Create a conda environment:
conda create -n n2n_SEM python=3.7
- Activate the environment:
source activate n2n_SEM
- Install packages:
pip install csbdeep
conda install tensorflow-gpu==1.15 # requires a sufficently new NVIDIA driver
pip install jupyter
- Start jupyter server:
jupyter notebook
The Noise2Noise training method was first described by Lehtinen et al.:
@article{lehtinen2018noise2noise,
title={Noise2noise: Learning image restoration without clean data},
author={Lehtinen, Jaakko and Munkberg, Jacob and Hasselgren, Jon and Laine, Samuli and Karras, Tero and Aittala, Miika and Aila, Timo},
journal={arXiv preprint arXiv:1803.04189},
year={2018}
}
The training framework used in this example is CSBDeep, which builds on Tensorflow.