tiff to omezarr and nrrd to precomputed converters by ddelpiano · Pull Request #6 · MetaCell/neuroglancer-python

ddelpiano · 2026-01-21T09:42:05Z

No description provided.

Copilot

Pull request overview

This PR adds two converter tools for preparing 3D image data for visualization in Neuroglancer: a TIFF to OME-Zarr converter and an NRRD to Neuroglancer Precomputed converter. Both converters support volumetric data processing, multichannel handling, and mesh generation for segmentation visualization.

Changes:

Added TIFF to OME-Zarr converter with multichannel support and diagnostic tools
Added NRRD to Precomputed converter with mesh generation and segment property management
Included utility scripts for data inspection, validation, and troubleshooting

Reviewed changes

Copilot reviewed 11 out of 12 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
converters/tif_to_omezarr/diagnose_tiff.py	Diagnostic tool for visualizing TIFF files and checking for black image issues
converters/tif_to_omezarr/converter.py	Main converter for TIFF to OME-Zarr with multichannel support and pyramid generation
converters/tif_to_omezarr/check_tiff_stats.py	Utility to inspect TIFF metadata and data statistics
converters/tif_to_omezarr/README.md	Documentation for TIFF to OME-Zarr converter usage
converters/nrrd_to_precomputed/meshes_generator_parallel.py	Parallel mesh generation for precomputed segmentation volumes
converters/nrrd_to_precomputed/meshes_generator.py	Mesh generation with physical coordinate transformation and merging options
converters/nrrd_to_precomputed/merge_segments.py	Tool for merging fragmented segments using connected component analysis
converters/nrrd_to_precomputed/inspect_nrrd.py	Utility to inspect NRRD file contents and identify segmentation issues
converters/nrrd_to_precomputed/converter.py	Main converter for NRRD to Neuroglancer Precomputed format
converters/nrrd_to_precomputed/check_mesh_coords.py	Debug tool for verifying mesh and volume coordinate spaces
converters/nrrd_to_precomputed/README.md	Documentation for NRRD to Precomputed workflow

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-21T09:42:54Z

+
+    zarrdir = Path(args.output_zarr_ome_dir)
+    if zarrdir.suffix != ".zarr":
+        print("Name of ouput zarr directory must end with '.zarr'")


Corrected spelling of 'ouput' to 'output'.

Suggested change

print("Name of ouput zarr directory must end with '.zarr'")

print("Name of output zarr directory must end with '.zarr'")

Copilot · 2026-01-21T09:42:54Z

+            print(f"Successfully wrote {ds_name}")
+            print(f"  Encoding: {encoding}")
+            print(
+                f"  Compression: {'gzip (. gz files)' if compress else 'none (raw files)'}"


Extra space in '. gz' should be '.gz'.

Suggested change

f" Compression: {'gzip (. gz files)' if compress else 'none (raw files)'}"

f" Compression: {'gzip (.gz files)' if compress else 'none (raw files)'}"

Copilot · 2026-01-21T09:42:54Z

+        if not is_planar_ome:
+            if keep_channels:
+                print("Image is single-channel")
+            if channel is not None and channel != 0 and not is_planar_ome:


The condition not is_planar_ome is checked twice in this block (lines 355 and 358). Consider restructuring to avoid redundancy.

Suggested change

if channel is not None and channel != 0 and not is_planar_ome:

if channel is not None and channel != 0:

Copilot · 2026-01-21T09:42:55Z

+        num_workers = min(
+            4, max(1, cpu_count() - 2)
+        )  # Use max 4 workers, leave 2 cores free


Magic numbers 4 and 2 should be defined as named constants for better maintainability.

seankmartin

Thanks for the PR, I didn't look at the conversion steps or data formats in detail, I trust that's what has to happen for the input/output. It's probably easier to visually debug these anyway. Just left a few comments about the ome-zarr + version, and a few specifics about cloud volume and meshes.

seankmartin · 2026-05-11T08:20:36Z

+    return {
+        'multiscales': [
+            {
+                'version': '0.4',


What is our reasoning for converting to ome-zarr 0.4 with zarr v2 instead of ome-zarr 0.5 with zarr v3? It's not a big deal since neuroglancer supports both, but if we're early in the process it might be good for us to consider to convert to the newer format instead

seankmartin · 2026-05-11T08:49:31Z

+
+    # Create dataset (using create_array for Zarr v3 compatibility)
+    # Use default compression which works across zarr versions
+    dataset = root.create_array(


If we do use ome-zarr 0.5, I think here is where we put the zarr_format arg for v3 - see https://zarr.readthedocs.io/en/stable/api/zarr/api/synchronous/#zarr.api.synchronous.create_array

seankmartin · 2026-05-11T09:02:51Z

+    resolution = vol.info["scales"][0]["resolution"]
+    voxel_offset = vol.info["scales"][0].get("voxel_offset", [0, 0, 0])
+
+    # Create mesh info with legacy format (multi-res format not supported by CloudVolume)


You can make a multi-res mesh with CloudVolume but it involves a bit of a different process. If you look in the cryoet_data_portal_neuroglancer repo at precompute/mesh.py you can see example

If the seg volumes are pretty small, and so the meshes not very detailed, the legacy format should be fine though

I'd suggest to remove the comment that it's not supported though, and replace that we choose to use legacy

seankmartin · 2026-05-11T09:11:35Z

+        print(f"    Voxel offset: {voxel_offset}")
+
+
+def generate_single_mesh(args):


What is here seems reasonable, just wanted to mention that neuroglancer python also has a volume creator if of interest to try, something like:

vol = neuroglancer.LocalVolume(data=volume_data, dimensions=coordinate_space) # this auto creates meshes for each unique value in the volume ids = [int(i) for i in np.unique(transposed_mesh[:])] // These can be directly written to legacy format with something like json_descriptor = '{{"fragments": ["mesh.{}.{}"]}}' for id in ids[1:]: mesh_data = vol.get_object_mesh(id) with open(str(mesh_path / ".".join(("mesh", str(id), str(id)))), "wb") as mesh_file: mesh_file.write(mesh_data) with open(str(mesh_path / "".join((str(id), ":0"))), "w") as frag_file: frag_file.write(json_descriptor.format(id, id))

seankmartin · 2026-05-11T09:20:17Z

+        return {"status": "error", "seg_id": seg_id, "error": str(e)}
+
+
+def generate_meshes_with_skimage(


Not sure how general purpose this is meant to be vs for a specific use case. If we're intending to reuse this it might be more useful to have meshes_generator_parallel import the functions from meshes_generator to have one source of truth. But at a glance I can't quite tell if there are subtle modifications to the functions for the purpose of the parallel processing

seankmartin · 2026-05-11T09:22:07Z

+        ]
+    )
+
+    zattrs_dict = {


Same comment as other file as to whether we need 0.4 for a reason

seankmartin · 2026-05-11T09:23:33Z

+    tzarr = zarr.open(
+        str(zarrdir),
+        mode="w",
+        zarr_format=2,


if we use ome zarr v0.5 this would need to change to zarr_format=3

tiff to omezarr and nrrd to precomputed converters

e64cda1

ddelpiano requested a review from Copilot January 21, 2026 09:42

Copilot AI reviewed Jan 21, 2026

View reviewed changes

first poc dm3 to zarr conversion

aabdae6

ddelpiano requested review from seankmartin and tarelli May 7, 2026 22:20

add script to convert spliceosome data

c9da690

seankmartin reviewed May 11, 2026

View reviewed changes

	print("Name of ouput zarr directory must end with '.zarr'")
	print("Name of output zarr directory must end with '.zarr'")

	f" Compression: {'gzip (. gz files)' if compress else 'none (raw files)'}"
	f" Compression: {'gzip (.gz files)' if compress else 'none (raw files)'}"

	if channel is not None and channel != 0 and not is_planar_ome:
	if channel is not None and channel != 0:

		print(f" Voxel offset: {voxel_offset}")


		def generate_single_mesh(args):

		return {"status": "error", "seg_id": seg_id, "error": str(e)}


		def generate_meshes_with_skimage(

Conversation

ddelpiano commented Jan 21, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

seankmartin left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

seankmartin May 11, 2026

Choose a reason for hiding this comment

Uh oh!

seankmartin May 11, 2026

Choose a reason for hiding this comment

Uh oh!

seankmartin May 11, 2026

Choose a reason for hiding this comment

Uh oh!

seankmartin May 11, 2026

Choose a reason for hiding this comment

Uh oh!

seankmartin May 11, 2026

Choose a reason for hiding this comment

Uh oh!

seankmartin May 11, 2026

Choose a reason for hiding this comment

Uh oh!

seankmartin May 11, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

seankmartin left a comment •

edited

Loading