Interactive parameter tuning with napari and magicgui#

A common task when processing image is to manually tune parameters until a segementation algorithm works fine on a give dataset. The napari viewer and magicgui offer the possibility to automatically build a user-interface where the user can tune parameters and see immediately how a corresponding segmentation will look like. This notebook demonstrates how this can be achieved from a simple python function that does the image processing.

# Inspired by
import napari
import pyclesperanto_prototype as cle
from magicgui import magicgui
from napari.types import ImageData, LabelsData

Our starting point is a function that implements a certain image processing workflow. In this example, we blur an image using a Gaussian filter, we threshold it using Otsu’s method and we apply connected component labeling to differentiate objects and label them.

def process_image(image_input: ImageData, sigma: float = 2) -> LabelsData:
    if input is not None:
        # process the image
        blurred = cle.gaussian_blur(image_input, sigma_x=sigma, sigma_y=sigma)
        binary = cle.threshold_otsu(blurred)
        labels = cle.connected_components_labeling_box(binary)

        return labels

We also need an example image to demonstrate the procedure.

# load data
from import imread
image = imread('')

Next, we start up the napari viewer, add the original image and append a automatically generated user interface, a dock widget, that allows us to configure the parameters of the function provided above. Furthermore, we take a screenshot of the viewer.

# start up napari
viewer = napari.Viewer()
viewer.add_image(image, name='blobs')

# generate a Graphical User Interface from the function above magically

# take a screenshot
Warning: Could not find scikit-tensor which is needed for separable approximations...
If you want to compute separable approximations, please install it with
pip install scikit-tensor-py3