pyclesperanto-prototype 0.18.0

GPU-accelerated image processing in python using OpenCL

py-clesperanto

py-clesperanto is a prototype for clesperanto - a multi-platform multi-language framework for GPU-accelerated image processing. We mostly use it in the life sciences for analysing 3/4-dimensional microsopy data, e.g. as we face it developmental biology when segmenting cells and studying their individual properties as well as properties of compounds of cells forming tissues.

Image data source: Daniela Vorkel, Myers lab, MPI-CBG, rendered using napari

clesperanto uses OpenCL kernels from CLIJ. Since version 0.11.1 py-clesperanto comes with a yet experimental cupy-based CUDA backend.

For users convenience, there are code generators available for napari and Fiji. Also check out the napari workflow optimizer for semi-automatic parameter tuning of clesperanto-functions.

Reference

The full reference is available as part of the CLIJ2 documentation.

Installation

• Get a conda/python environment, e.g. via mini-conda. If you never used python/conda environments before, please follow the instructions here first.

conda create --name cle_39 python=3.9
conda activate cle_39

• Install pyclesperanto-prototype using conda:

conda install -c conda-forge pyclesperanto-prototype

OR using pip:

pip install pyclesperanto-prototype

Mac-users please also install this:

conda install -c conda-forge ocl_icd_wrapper_apple

Linux users please also install this:

conda install -c conda-forge ocl-icd-system

Example code

A basic image procressing workflow loads blobs.gif and counts the number of gold particles:

import pyclesperanto_prototype as cle

from skimage.io import imread, imsave

# initialize GPU
device = cle.select_device("GTX")
print("Used GPU: ", device)

# load data
image = imread('https://imagej.nih.gov/ij/images/blobs.gif')

# process the image
inverted = cle.subtract_image_from_scalar(image, scalar=255)
blurred = cle.gaussian_blur(inverted, sigma_x=1, sigma_y=1)
binary = cle.threshold_otsu(blurred)
labeled = cle.connected_components_labeling_box(binary)

# The maxmium intensity in a label image corresponds to the number of objects
num_labels = cle.maximum_of_all_pixels(labeled)

# print out result
print("Num objects in the image: " + str(num_labels))

# save image to disc
imsave("result.tif", labeled)

Example gallery

	Select GPU
	Image processing in Jupyter Notebooks
	Counting blobs
	Voronoi-Otsu labeling
	3D Image segmentation
	Cell segmentation based on membranes
	Counting nuclei according to expression in multiple channels
	Differentiating nuclei according to signal intensity
	Detecting beads and measuring their size
	Label statistics
	Parametric maps
	Crop and paste images
	Inspecting 3D image data
	Rotation, scaling, translation, affine transforms
	Deskewing
	Multiply vectors and matrices
	Matrix multiplication
	Working with spots, pointlist and matrices Lists of nonzero pixel coordinates
	Mesh between centroids
	Mesh between touching neighbors
	Mesh with distances
	Mesh nearest_neighbors
	Export to igraph and networkx
	Neighborhood definitions
	Tissue neighborhood quantification
	Neighbors of neighbors
	Voronoi diagrams
	Shape descriptors based on neighborhood graphs
	Measuring distances between labels in two label images
	Tribolium morphometry + Napari
	Tribolium morphometry (archived version)
	napari+dask timelapse processing

Technical insights

	Browsing operations
	Interactive widgets
	Automatic workflow optimization
	Tracing memory consumtion on NVidia GPUs
	Exploring and switching between GPUs
	Interoperability with cupy Using the cupy backend

Related projects

	napari-pyclesperanto-assistant: A graphical user interface for general purpose GPU-accelerated image processing and analysis in napari.
	napari-accelerated-pixel-and-object-classification: GPU-accelerated Random Forest Classifiers for pixel and labeled object classification
	napari-clusters-plotter: Clustering of objects according to their quantitative properties

Benchmarking

We implemented some basic benchmarking notebooks allowing to see performance differences between pyclesperanto and some other image processing libraries, typically using the CPU. Such benchmarking results vary heavily depending on image size, kernel size, used operations, parameters and used hardware. Feel free to use those notebooks, adapt them to your use-case scenario and benchmark on your target hardware. If you have different scenarios or use-cases, you are very welcome to submit your notebook as pull-request!

• Affine transforms
• Background subtraction
• Gaussian blur
• Convolution
• Otsu's thresholding
• Connected component labeling
• Extend labels
• Statistics of labeled pixels / regionprops
• Histograms
• Matrix multiplication
• Pixel-wise comparison
• Intensity projections
• Axis transposition
• Nonzero

See also

There are other libraries for code acceleration and GPU-acceleration for image processing.

• numba
• cupy
• cucim
• clij

Feedback welcome!

clesperanto is developed in the open because we believe in the open source community. See our community guidelines. Feel free to drop feedback as github issue or via image.sc