Fix class_weights background voxel count to use actual array volume size

[Copilot]

class_weights was computing total_voxels = sum(fg_counts.values()) — the sum of all foreground voxels across all classes — rather than the actual spatial volume of each class's data array. This made bg = total_voxels - fg incorrect, corrupting the fg/bg imbalance ratio used for weighted sampling.

Changes

• CellMapImage.total_voxels — new property; product of the s0 array's spatial shape, scaled to training resolution via the existing _scale_count helper
• EmptyImage.total_voxels — returns 0 (no backing array)
• CellMapDataset.class_counts — adds "totals_total" key alongside "totals", sourced from each target image's total_voxels
• CellMapMultiDataset.class_counts — aggregates "totals_total" across datasets
• CellMapMultiDataset.class_weights — computes bg = totals_total[cls] - fg per class instead of summing across all fg counts

# Before (wrong): bg was total fg across all classes minus this class's fg
total_voxels = sum(counts.values())  # sum of fg voxels for all classes
bg = total_voxels - fg

# After (correct): bg is the actual array volume minus this class's fg
bg = max(total_counts[cls] - fg, 0)  # total_counts[cls] = np.prod(array_shape) scaled to training res

---

📍 Connect Copilot coding agent with Jira, Azure Boards or Linear to delegate work to Copilot in one click without leaving your project management tool.

[rhoadesScholar]

@copilot Address the purpose of this PR

[Copilot]

Copilot wasn't able to review any files in this pull request.

[codecov]

Codecov Report

❌ Patch coverage is 81.25000% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.22%. Comparing base (ce2a150) to head (ef33153).
⚠️ Report is 9 commits behind head on rewrite.

Files with missing lines	Patch %	Lines
src/cellmap_data/image.py	69.23%	4 Missing ⚠️
src/cellmap_data/dataset.py	80.00%	1 Missing ⚠️
src/cellmap_data/empty_image.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           rewrite      #66      +/-   ##
===========================================
+ Coverage    80.81%   81.22%   +0.41%     
===========================================
  Files           23       23              
  Lines         1538     1561      +23     
===========================================
+ Hits          1243     1268      +25     
+ Misses         295      293       -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR updates how class voxel statistics are computed across CellMapImage → CellMapDataset → CellMapMultiDataset, adding explicit “total voxel” counts so class sampling weights can use true background volume rather than deriving totals from foreground counts.

Changes:

• Add CellMapImage.total_voxels and propagate per-class total-volume counts via a new totals_total key in class_counts.
• Update CellMapMultiDataset.class_weights to compute background as totals_total - totals per class.
• Extend unit tests to validate the new keys and the corrected weight calculation.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/cellmap_data/image.py	Adds total_voxels for computing per-image volume at training resolution.
src/cellmap_data/empty_image.py	Adds a total_voxels stub for API compatibility with EmptyImage.
src/cellmap_data/dataset.py	Extends class_counts to include totals_total per class.
src/cellmap_data/multidataset.py	Aggregates totals_total and fixes class_weights background calculation.
tests/test_image.py	Adds a test asserting total_voxels equals array size in the fixture.
tests/test_multidataset.py	Adds tests for totals_total and for corrected background/weight math.

[Copilot]

@copilot Address the purpose of this PR

The PR's purpose is to fix total_voxels to reflect the actual data volume size (spatial shape) for each class, rather than using the sum of all foreground voxel counts. CellMapImage.total_voxels is now derived from the cached bounding_box (world-space extent divided by training-resolution voxel size) so no extra zarr I/O is needed. CellMapDataset.class_counts exposes a "totals_total" key, and class_weights computes bg = totals_total[cls] - fg per class. Committed in 0d4bf5a.

[Copilot]

Pull request overview

This PR refines how class statistics are computed for sampling by distinguishing per-class foreground voxel counts from the total voxel volume of the underlying label arrays, ensuring background-vs-foreground ratios are based on the real volume size rather than inferred totals.

Changes:

• Extend class_counts to return both per-class foreground totals (totals) and per-class total volume (totals_total) across CellMapDataset and CellMapMultiDataset.
• Update CellMapMultiDataset.class_weights to compute background voxels using totals_total - totals.
• Add CellMapImage.total_voxels (and EmptyImage.total_voxels) and introduce targeted regression tests for the new behavior.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/cellmap_data/image.py	Adds total_voxels derived from bounding_box and training scale.
src/cellmap_data/empty_image.py	Provides total_voxels for empty sources (returns 0).
src/cellmap_data/dataset.py	Expands class_counts to include totals_total and caches the result.
src/cellmap_data/multidataset.py	Aggregates totals_total across datasets and fixes class_weights background calculation.
tests/test_image.py	Adds a test validating total_voxels equals array size for a simple fixture.
tests/test_multidataset.py	Adds tests asserting totals_total and class weight math use the true volume size.