Alexander Krull^1,2, Tim-Oliver Buchholz², Florian Jug
¹[email protected], ²Authors contributed equally

The field of image denoising is currently dominated by discriminative deep learning methods that are trained on pairs of noisy input and clean target images. Recently it has been shown that such methods can also be trained without clean targets. Instead, independent pairs of noisy images can be used, in an approach known as NOISE2NOISE (N2N). Here, we introduce NOISE2VOID (N2V), a training scheme that takes this idea one step further. It does not require noisy image pairs, nor clean target images. Consequently, N2V allows us to train directly on the body of data to be denoised and can therefore be applied when other methods cannot. Especially interesting is the application to biomedical image data, where the acquisition of training targets, clean or noisy, is frequently not possible. We compare the performance of N2V to approaches that have either clean target images and/or noisy image pairs available. Intuitively, N2V cannot be expected to outperform methods that have more information available during training. Still, we observe that the denoising performance of NOISE2VOID drops in moderation and compares favorably to training-free denoising methods.

Paper: https://arxiv.org/abs/1811.10980

Our implementation is based on CSBDeep (github).

Eva Höck^1,⚹, Tim-Oliver Buchholz^2,⚹, Anselm Brachmann^1,⚹, Florian Jug^3,⁜, and Alexander Freytag^1,⁜
1Carl Zeiss AG, Germany
²Facility for Advanced Imaging and Microscopy, Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland
³Jug Group, Fondazione Human Technopole, Milano, Italy
^{⚹, ⁜}Equal contribution

In recent years, neural network based image denoising approaches have revolutionized the analysis of biomedical microscopy data. Self-supervised methods, such as Noise2Void (N2V), are applicable to virtually all noisy datasets, even without dedicated training data being available. Arguably, this facilitated the fast and widespread adoption of N2V throughout the life sciences. Unfortunately, the blind-spot training underlying N2V can lead to rather visible checkerboard artifacts, thereby reducing the quality of final predictions considerably. In this work, we present two modifications to the vanilla N2V setup that both help to reduce the unwanted artifacts considerably. Firstly, we propose a modified network architecture, i.e. using BlurPool instead of MaxPool layers throughout the used UNet, rolling back the residual-UNet to a non-residual UNet, and eliminating the skip connections at the uppermost UNet level. Additionally, we propose new replacement strategies to determine the pixel intensity values that fill in the elected blind-spot pixels. We validate our modifications on a range of microscopy and natural image data. Based on added synthetic noise from multiple noise types and at varying amplitudes, we show that both proposed modifications push the current state-of-the-art for fully self-supervised image denoising.

OpenReview: https://openreview.net/forum?id=IZfQYb4lHVq

This implementation requires Tensorflow. We have tested Noise2Void using Python 3.9 and TensorFlow 2.7, 2.10 and 2.13.

⚠️ n2v is not compatible with TensorFlow 2.16 ⚠️

Note: If you want to use TensorFlow 1.15 you have to install N2V v0.2.1. N2V v0.3.* supports TensorFlow 2 only.

We recommend using miniconda. If you do not yet have a strong opinion, just use it too!

After installing Miniconda, create a conda environment:

conda create -n 'n2v' python=3.9
conda activate n2v

Since TensorFlow 2.16, n2v is no longer compatible with the latest tensorflow version. Therefore we recommend installing the latest versions with which it was tested. The following instructions are copied from TensorFlow website, using the WayBack machinbe.

Linux:

conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_PREFIX/lib/
python3 -m pip install tensorflow

macOs:

# There is currently no official GPU support for MacOS.
python3 -m pip install tensorflow

Windows

# Last supported Windows with GPU without WSL2
conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
python3 -m pip install tensorflow

Linux:

conda install -c conda-forge cudatoolkit=11.8.0
python3 -m pip install nvidia-cudnn-cu11==8.6.0.163 tensorflow==2.13.*
mkdir -p $CONDA_PREFIX/etc/conda/activate.d
echo 'CUDNN_PATH=$(dirname $(python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"))' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export LD_LIBRARY_PATH=$CUDNN_PATH/lib:$CONDA_PREFIX/lib/:$LD_LIBRARY_PATH' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
source $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh

macOs:

# There is currently no official GPU support for MacOS.
python3 -m pip install tensorflow

Windows WSL2

conda install -c conda-forge cudatoolkit=11.8.0
python3 -m pip install nvidia-cudnn-cu11==8.6.0.163 tensorflow==2.13.*
mkdir -p $CONDA_PREFIX/etc/conda/activate.d
echo 'CUDNN_PATH=$(dirname $(python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"))' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export LD_LIBRARY_PATH=$CUDNN_PATH/lib:$CONDA_PREFIX/lib/:$LD_LIBRARY_PATH' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
source $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh

Linux:

python3 -m pip install tensorflow[and-cuda]

macOs:

# There is currently no official GPU support for MacOS.
python3 -m pip install tensorflow

Windows WSL2

python3 -m pip install tensorflow[and-cuda]

$ pip install n2v

This option is ideal if you want to edit the code. Clone the repository:

$ git clone https://github.com/juglab/n2v.git

Change into its directory and install it:

$ cd n2v
$ pip install -e .

You are now ready to run Noise2Void.

Have a look at our jupyter notebook:

• 2D example BSD68
• 2D example SEM
• 2D example RGB
• 3D example
• 2D StructN2V example synth_mem

In order to run the notebooks, install jupyter in your conda environment:

pip install jupyter

Coming soon:

• N2V2 example notebooks.

Note: You can use the N2V2 functionality by providing the following three parameters to the N2V-Config object:

• blurpool=True, by default set to False
• skip_skipone=True, by default set to False
• n2v_manipulator="median", by default set to "uniform_withCP"
• unet_residual=False, by default set to False

Warning: Currently, N2V2 does only support 2D data.
Warning: We have not tested N2V2 together with struct-N2V.

N2V, N2V2 and structN2V are available in napari!

@inproceedings{krull2019noise2void,
  title={Noise2void-learning denoising from single noisy images},
  author={Krull, Alexander and Buchholz, Tim-Oliver and Jug, Florian},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={2129--2137},
  year={2019}
}
N2V2 citation coming soon.

see here for more info on StructN2V.

The functional "tests" are meant to be run as regular programs. They are there to make sure that examples are still running after changes.

In some cases tensorflow is unable to allocate GPU memory and fails. One possible solution could be to set the following environment variable: export TF_FORCE_GPU_ALLOW_GROWTH=true