GPU-accelerated image processing in napari using OpenCL
The py-clEsperanto-assistant is a yet experimental napari plugin for building GPU-accelerated image processing workflows. It is part of the clEsperanto project and thus, aims at removing programming language related barriers between image processing ecosystems in the life sciences. It uses pyclesperanto and with that pyopencl as backend for processing images. This plugin was generated with Cookiecutter using with napari's cookiecutter-napari-plugin template.
It is recommended to install the assistant via conda:
conda create --name bio11 python==3.8.5 conda activate bio11 conda install -c conda-forge pyopencl==2021.2.1 pip install napari-pyclesperanto-assistant pip install "napari[all]"
Alternatively, you can install the assistant using napari's plugin installer in the menu
Plugins > Install/uninstall Packages.
Windows users should paste this URL
in this field and click on
Install before proceeding:
Afterwards, click install clEsperanto like by clicking on
You can then start napari, e.g. from command line, and find the assistant in the
pyclesperanto offers various possibilities for processing images. It comes from developers who work in life sciences and thus, it may be focused towards processing two- and three-dimensional microscopy image data showing cells and tissues. A selection of pyclesperanto's functionality is available via the assistant user interface. Typical workflows which can be built with this assistant include
- image filtering
- denoising / noise reduction (mean, median, Gaussian blur)
- background subtraction for uneven illumination or out-of-focus light (bottom-hat, top-hat, subtract Gaussian background)
- grey value morphology (local minimum, maximum. variance)
- gamma correction
- Laplace operator
- Sobel operator
- combining images
- image math (adding, subtracting, multiplying, dividing images)
- absolute / squared difference
- image transformations
- reduce stack
- image projections
- minimum / mean / maximum / sum / standard deviation projections
- image segmentation
- binarization (thresholding, local maxima detection)
- instance segmentation
- semantic segmentation
- detect label edges
- label spots
- connected component labeling
- post-processing of binary images
- binary opening
- binary closing
- binary and / or / xor
- post-processing of label images
- dilation (expansion) of labels
- extend labels via Voronoi
- exclude labels on edges
- exclude labels within / out of size / value range
- merge touching labels
- parametric maps
- proximal / touching neighbor count
- distance measurements to touching / proximal / n-nearest neighbors
- pixel count map
- mean / maximum / extension ratio map
- label measurements / post processing of parametric maps
- minimum / mean / maximum / standard deviation intensity maps
- minimum / mean / maximum / standard deviation of touching / n-nearest / neighbors
- neighbor meshes
- touching neighbors
- n-nearest neighbors
- proximal neighbors
- distance meshes
- measurements based on label images
- bounding box 2D / 3D
- minimum / mean / maximum / sum / standard deviation intensity
- center of mass
- mean / maximum distance to centroid (and extension ratio shape descriptor)
- mean / maximum distance to center of mass (and extension ratio shape descriptor)
- code export
- python / Fiji-compatible jython
- python jupyter notebooks
- pyclesperanto scripting
- cell segmentation
- cell counting
- cell differentiation
- tissue classification
Start up the assistant
Start up napari, e.g. from the command line:
Load example data, e.g. from the menu
File > Open Samples > clEsperanto > CalibZAPWfixed and
start the assistant from the menu
Plugins > clEsperanto > Assistant. Select a GPU in case you are asked to.
In case of two dimensional timelapse data, an initial conversion step might be necessary depending on your data source.
Click the menu
Plugins > clEsperanto > Convert to 2d timelapse. In the dialog, select the dataset and click ok.
You can delete the original dataset afterwards:
Set up a workflow
Choose categories of operations in the top right panel, for example start with denoising using a Gaussian Blur with sigma 1 in x and y.
Continue with background removal using the top-hat filter with radius 5 in x and y.
For labeling the objects, use Voronoi-Otsu-Labeling with both sigma parameters set to 2.
The labeled objects can be extended using a Voronoi diagram to derive a estimations of cell boundaries.
You can then configure napari to show the label boundaries on top of the original image:
When your workflow is set up, click the play button below your dataset:
You can also export your workflow as Python/Jython code or as notebook.
After exporting your workflow as Jupyter notebook, you can start the notebook from the command line using
jupyter notebook my_notebook.ipynb
In some cases you need to replace the command
cle.imread('None)` with a command loading your image data.
After that, you can execute the notebook.
Also note: The generated python/jython code is not capable of processing timelapse data, you need to program a for-loop processing timepoints individually yourself.
Work in progress, contributions welcome.
Getting the recent code from github and locally installing it
git clone https://github.com/clesperanto/napari_pyclesperanto_assistant.git pip install -e ./napari_pyclesperanto_assistant
Optional: Also install pyclesperantos recent source code from github:
git clone https://github.com/clEsperanto/pyclesperanto_prototype.git pip install -e ./pyclesperanto_prototype
- 21 October 2021
- 19 December 2020
- 3 - Alpha
- Stars: 12
- Forks: 4
- Issues + PRs: 3
Python versions supported:
- OS Independent
- napari-plugin-engine (>=0.1.4)
- napari (>=0.4.7)
- pyclesperanto-prototype (>=0.10.0)
- numpy (!=1.19.4)