[BUG] Cell candidates do not scale properly in napari viewer

[grinic]

Describe the bug
Hi, thanks for the fantastic work and software! I hope I am not doing anything wrong, but I noticed this bug in the napari plugin.

Cell candidates layer generated with the napari plugin (cellfinder -> cell detection) do not scale corretcly.

To Reproduce
Add an image to the napari viewer setting anisotropic scale:

import napari
from skimage.io import imread

v = napari.Viewer()

signal = imread("path\to\signal.tiff")
background = imread("path\to\background.tiff")

v.add_image(background,
            name          = 'ch:0-488-autofluo',
            scale           = (5, 1.46, 1.46),
            colormap        = 'green',
            interpolation2d = 'nearest',
            blending        = 'additive',
           )

v.add_image(signal,
            name          = 'ch:2-642-iDisco',
            scale           = (5, 1.46, 1.46),
            colormap        = 'magenta',
            interpolation2d = 'nearest',
            blending        = 'additive',
           )

Run the cellfinder napari plugin for cell detection setting anisotropic voxel size

Voxel size (z) : 5
Voxel size (y) : 1.46
Voxel size (x) : 1.46

Observe that the points in the Cell candidates layer do not overlap correctly with the image.

Screenshots

However, when the candidates are added manually with

candidates = v.layers['Cell candidates'].data

v.add_points(
    candidates,
    name='Cell candidates scaled',
    symbol='ring',
    blending='additive',
    face_color = 'lightskyblue',
    border_color = 'transparent',
    out_of_slice_display  = True,
    scale = (5, 1.46, 1.46)
)

the points overlap properly:

Computer used (please complete the following information):

• OS: Windows 10
• Version cellfinder: 1.8.0
• Hardware specs 256 GB RAM, Xeon gold CPU (8 cores).

[adamltyson]

Thanks for raising this @grinic. I think the plugin doesn't "know" anything about the image scaling, so doesn't add this to the resulting cells. We may have built this plugin before napari had the concept of scaling!

Anyway, we should add this, thanks for letting us know.

In the meantime, it should all work ok if you just don't scale your input data.

[AlgoFoe]

Hi @adamltyson , I would be happy to take a first pass at fixing this.
From what I understand, the napari plugin isn’t propagating the image scale to the generated points layer, so the cell candidate coordinates are correct in data space but not displayed correctly when anisotropic scaling is applied.

I will try to reproduce this locally and work on a fix.