ExtractChannel and FlatFieldDatasource 5D / multichannel datasource fixes

[barentine]

Addresses issues:

• ExtractChannel._pickChannel uses legacy image.data access and collapses the wrong index if you pass it a XYZTC shape [1024, 1024, 1, 1, 4] datasource and try to pick out one of the 4 channels
• FlatFieldDataSource doesn't preserve parent datasource t/z info in its current state, and can also mess up channel extraction.

Is this a bugfix or an enhancement?
bugfix
Proposed changes:

• use image.data_xyztc access in ExtractChannel module
• add units tests for ExtractChannel
• store parent datasource t/z info in FlatFieldDataSource

FlatfieldDataSource currently inherits from XYTCDataSource, i.e. the old-style datasource base. Certainly it would be better to do a full re-jig, but this was the fastest fix for me.

Checklist:
The below is a list of things what will be considered when reviewing PRs. It is not prescriptive, and does not
imply that PRs which touch any of these will be rejected but gives a rough indication of where there is a potential
for hangups (i.e. factors which could turn a 5 min review into a half hour or longer and shunt it to the bottom
of the TODO list).

• Does the PR avoid variable renaming in existing code, whitespace changes, and other forms of tidying? [There is a place for code tidying, but it makes reviewing
  much simpler if this is kept separate from functional changes. The auto-formatting performed by some editors is particulaly egregious and can lead to files with thousands
  of non-functional changes with a few functional changes scattered amoungst them]

If an enhancement (or non-trivial bugfix):

• Has this been discussed in advance (feature request, PR proposal, email, or direct conversation)?
• Does this change how users interact with the software? How will these changes be communicated?
• Does this maintain backwards compatibility with old data?
• Does this change the required dependencies?
• Are there any other side effects of the change?

[David-Baddeley]

We should really go a step further and try and make this lazy (if practical for the dataset). Otherwise we're going to be forcing lots of stuff into memory un-necessarily.

[barentine]

I'll have a go at making it lazy, though I don't love letting perfection be the enemy of bug fixing ;)

[David-Baddeley]

Sorry, I didn't mean it had to happen. Before merging ... Was just what I thought when looking at it.

…

On Sat, 25 Apr 2026, 07:51 Andrew E S Barentine, ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In PYME/recipes/base.py <#1654 (comment)> : > @@ -1021,7 +1021,11 @@ class ExtractChannel(ModuleBase): channelToExtract = Int(0) def _pickChannel(self, image): - chan = image.data[:,:,:,self.channelToExtract] + c = self.channelToExtract + # Use data_xyztc (5D, X,Y,Z,T,C) with a slice (c:c+1) rather than an integer index + # so the channel axis is preserved, avoiding collapse to a 4D array and re-promotion ambiguity. + # chan = image.data_xyztc[:,:,:,:,self.channelToExtract] # fails to preserve channel axis + chan = image.data_xyztc[:,:,:,:,c:c+1] im = ImageStack(chan, titleStub = 'Filtered Image') I'll have a go at making it lazy, though I don't love letting perfection be the enemy of bug fixing ;) — Reply to this email directly, view it on GitHub <#1654 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEUSWYALIBC3LX5AISNFOX34XPATNAVCNFSM6AAAAACWXLTDYCVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHM2DCNZTGAZDQNJTGA> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/AEUSWYHSTZBEQ7N4TK6ISWL4XPATNA5CNFSNUABKM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UKJSXM2LFO4XTIMJXGMYDEOBVGMYKM4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJKTGN5XXIZLSL5UW64Y> and Android <https://github.com/notifications/mobile/android/AEUSWYERFUOST7X5P2M2Q734XPATNA5CNFSNUABKM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UKJSXM2LFO4XTIMJXGMYDEOBVGMYKM4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJLTGN5XXIZLSL5QW4ZDSN5UWI>. Download it today! You are receiving this because you commented.Message ID: ***@***.*** com>

[barentine]

No apology needed, David!

I added a color crop to CropDataSource, which looks like a lazy slicer, and made ExtractChannel just a color crop. Can back it out if you prefer an alternative.

[barentine]

I've added a change to move FlatFieldDataSource to inherit from the 5D XYZTCDataSource. @David-Baddeley let me know if you'd like this PR refactored into a couple smaller ones.

[barentine]

Note that the test failure appears unrelated (and passes on my machine). We probably want to look at whether test_fiducial_matching.py is doing what we want it to.