Request: ZARR-support

[macromeer]

Hi there,

the napari_pyclesperanto_assistant is a great plugin for teaching. I was wondering whether support for zarr data would be possible. Currently, after loading a zarr dataset using the napari-ome-zarr plugin, applying functions from the napari_pyclesperanto_assistant result in a TypeError.

For the time being, I will follow the dask-based approach.

[haesleinhuepf]

Hi @macromeer ,

that's a great suggestion!

result in a TypeError.

Could you share an example workflow and the exact error message? This might help to estimate the effort needed to implement this.

Thanks!

Best,
Robert

[macromeer]

Hi @haesleinhuepf - thanks for reacting so quickly! This should enable you to quickly reproduce the TypeError:

Create fresh environment:

conda create -y -n zarr-napari-env -c conda-forge python=3.10 # ome_zarr has bug if using python < 3.10
conda activate zarr-napari-env
python -m pip install "napari[all]"
pip install napari-ome-zarr
pip install napari-pyclesperanto-assistant

Run Python script in zarr-napari-env:

import napari
from ome_zarr.utils import download
remote_url = "https://uk1s3.embassy.ebi.ac.uk/idr/zarr/v0.4/idr0062A/6001240.zarr"
local_path = input("choose local path")
download(input_path=remote_path,output_dir=local_path) # zarr dataset needs to be local
viewer = napari.Viewer()
viewer.open(local_path, plugin="napari-ome-zarr")
napari.run()

When I use a function from the napari_pyclesperanto_assistant, I get for example

TypeError: _threshold_otsu(): incompatible function arguments. The following argument types are supported:
    1. (device: pyclesperanto._pyclesperanto._Device, src: pyclesperanto._pyclesperanto._Array, dst: pyclesperanto._pyclesperanto._Array) -> pyclesperanto._pyclesperanto._Array

Invoked with: (OpenCL) NVIDIA T400 4GB (OpenCL 3.0 CUDA)
	Vendor:                      NVIDIA Corporation
	Driver Version:              560.35.03
	Device Type:                 GPU
	Compute Units:               6
	Global Memory Size:          3696 MB
	Maximum Object Size:         924 MB
	Max Clock Frequency:         1425 MHz
	Image Support:               Yes
, <MultiScaleData at 0x7d7119368af0. 3 levels, 'uint16', shapes: ((236, 275, 271), (236, 137, 135), (236, 68, 67))>, None

[haesleinhuepf]

Ok, that's obviously an error within pyclespranto. Let's see if it can be fixed there 🌞

[StRigaud]

Hi @macromeer, @haesleinhuepf

I made a PR to add support of MultiScaleData type:

• Add-multiscaldata-compatibility pyclesperanto#306

and I provide also an example with a demo notebook, to be improved for better documentation. I went for the minimal integration for now as I am not very much accounted with the OME-Zarr and Napari mixup.

For info, it seems that MultiScaleData simply contains several dask.array.core.Array which are already supported by pycle, the PR simply identify the input being a MultiScaleData and take the highest resolution layer and feed it to pycle. The output return by pycle will loose the multiscale ressources and will be either of type pycle.Array or numpy.array.

Once pushed and deployed, there might be some extra work to do on the assistant, I did not test that.

Alternatively, we could consider that the assistant identify the data to be a MultiScaleData and can ask to the user the resolution to be process before sending the corresponding dask.array.core.Array to pycle. This approach might be cleaner or more flexible.

[StRigaud]

This functionnality was merged and will be available with the next release.
Best,

[StRigaud]

Hi @macromeer

The 0.16.1 is released, could you update your pyclesperanto version and try on your side if all is good?
Thanks in advance 😃

[macromeer]

Good timing, I've just tried it and it works 🥳 Thanks a lot! I will use it in my next teaching course.

Edit: Just fyi

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
napari-pyclesperanto-assistant 0.25.0 requires pyclesperanto==0.14.2, but you have pyclesperanto 0.16.1 which is incompatible.

Still works, though.

[StRigaud]

Good timing, I've just tried it and it works 🥳 Thanks a lot! I will use it in my next teaching course.

Great! Happy to help.

ERROR: pip's dependency

That's for @haesleinhuepf if he find 5 min 😆

[macromeer]

No dependency conflict in this fresh install 👍

mamba create --name devbio-napari-env python=3.9 -c conda-forge
mamba activate devbio-napari-env
mamba install devbio-napari -c conda-forge -c pytorch
pip install pyclesperanto==0.16.1 
pip install napari-ome-zarr

[macromeer]

Some operations that change type do result in an error though:

AttributeError: 'MultiScaleData' object has no attribute 'astype'
2025-02-18 11:11:53.116 | INFO     | napari_assistant._gui._category_widget:call_op:185 - threshold_otsu (pyclesperanto)(..., )
Segmentation fault (core dumped)

[StRigaud]

I don't have the issue outside of the assistant (see notebook demo) so my (quick) guess is that there is a type cast before calling threhsold_otsu in the assistant.

It's related to some code here:
https://github.com/haesleinhuepf/napari-assistant/blob/fc6c8e9769a8b7c1b3b2d840e35cc6a3a70e1ace/napari_assistant/_gui/_category_widget.py#L181-L191

I am not realy sure what's happening there but @haesleinhuepf should know I guess.

I reopen the issue until then