Fix multi-channel compositing and add per-channel norm support by timtreis · Pull Request #451 · scverse/spatialdata-plot

timtreis · 2025-04-22T22:10:18Z

Summary

Signal-based multi-channel compositing: replaces 4 inconsistent compositing paths in _render_images with a shared _additive_blend helper. For each channel, the "signal" (deviation from cmap(0)) is summed onto a canvas colored by the mean zero-value. This handles both Napari-style additive blending (black-to-color LUTs) and ImageJ Composite Invert (white-to-color LUTs), and produces reasonable results for arbitrary colormaps.
Per-channel normalization: render_images(norm=...) now accepts a list of Normalize objects for independent contrast control per channel (e.g. bright DAPI vs dim cytokine stain).
Removes dead percentiles_for_norm field and its deprecation alias.

Bug fixes

Bug	Fix
Averaging instead of additive sum in multi-cmap paths	Signal-based blend via `_additive_blend`
Missing clip to [0, 1] in 2-3 channel path	`np.clip` in `_additive_blend`
Double-alpha application in 4+ channel path (baked into RGBA and passed to imshow)	Alpha only via `imshow`
Redundant `isinstance(c, str)` filter in palette path	Removed (validated upstream)
White saturation with non-black-based cmaps (e.g. Reds, seismic)	Signal subtraction handles any cmap zero-value
`na_color` alpha compared via internal hex string	Uses `Color.get_alpha_as_float()` public API

Per-channel norm usage

from matplotlib.colors import Normalize

sdata.pl.render_images(
    "morphology_focus",
    palette=["#0f73e6", "#a4a400", "#f300a5", "#008a00"],
    channel=["DAPI", "18S", "ATP1A1/CD45/E-Cadherin", "AlphaSMA/Vimentin"],
    norm=[Normalize(0, 5000), Normalize(0, 2000), Normalize(0, 8000), Normalize(0, 3000)],
).pl.show()

Test plan

5 existing baseline images updated (signal-based output differs from averaging)
4 new visual tests with CI-generated baselines (5-channel compositing, per-channel norm)
3 error tests for norm validation (length mismatch, empty list, tuple rejection)

Closes #416, #534. Addresses #370. Partially addresses #460 (per-channel norm, but not arbitrary preprocessing callables).

Note: #450 (PCA aggregation) is not addressed by this PR.

timtreis · 2025-04-22T22:15:16Z

@Sonja-Stockhaus did we discuss this at one point? I've been ignoring that warning for a long time because it's thrown my matplotlib and doesn't fail

Sonja-Stockhaus · 2025-04-23T13:07:25Z

@Sonja-Stockhaus did we discuss this at one point? I've been ignoring that warning for a long time because it's thrown my matplotlib and doesn't fail

Nope. I also don't remember ever trying to render sth with 3+ channels

codecov-commenter · 2025-05-11T19:57:17Z

Codecov Report

❌ Patch coverage is 88.37209% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.50%. Comparing base (2893343) to head (11dbd79).

Files with missing lines	Patch %	Lines
src/spatialdata_plot/pl/utils.py	62.50%	1 Missing and 2 partials ⚠️
src/spatialdata_plot/pl/render.py	93.75%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #451      +/-   ##
==========================================
+ Coverage   73.68%   74.50%   +0.82%     
==========================================
  Files           9        9              
  Lines        2744     2750       +6     
  Branches      651      656       +5     
==========================================
+ Hits         2022     2049      +27     
+ Misses        447      419      -28     
- Partials      275      282       +7

Files with missing lines	Coverage Δ
src/spatialdata_plot/pl/basic.py	`82.29% <100.00%> (+0.06%)`	⬆️
src/spatialdata_plot/pl/render_params.py	`86.22% <100.00%> (ø)`
src/spatialdata_plot/pl/render.py	`86.44% <93.75%> (+2.44%)`	⬆️
src/spatialdata_plot/pl/utils.py	`65.71% <62.50%> (+0.41%)`	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

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

…nel norm Replace inconsistent compositing formulas in _render_images with a shared _additive_blend helper that implements standard additive blending with clamping, matching Napari/ImageJ/FIJI behavior. Fixes: - Averaging bug in paths 2A-cmap and 2D (divided by n_channels) - Missing clip in paths 2B (2-3ch) and 2C - Double-alpha bug in path 2B (4+ch) - Redundant isinstance(c, str) filter in path 2C Adds per-channel normalization support: render_images() now accepts norm as a list of Normalize objects (one per channel), enabling proper contrast control for multi-channel fluorescence data. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The `percentiles_for_norm` field on `ImageRenderParams` and its `quantiles_for_norm` deprecation alias were never wired into the rendering pipeline. Per-channel normalization is now handled via the `norm` parameter accepting a list of Normalize objects. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Updates 5 baseline images that changed due to the switch from averaging to additive blending in multi-channel image compositing. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The naive additive blend (sum RGB + clip) only works with black-to-color LUTs. For white-to-color cmaps (e.g. Reds, Greens) it saturates to white; for diverging cmaps (e.g. seismic) it washes out. Signal-based blending subtracts each cmap's zero-value before summing, then composites onto a canvas colored by the mean zero-value. This reduces to standard additive for black-to-color LUTs (Napari/ImageJ) and to invert-add-invert for white-to-color LUTs (ImageJ Composite Invert), while producing reasonable results for arbitrary cmaps. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- 5-channel auto-color composite (path 2B, n>3): untested path - 5-channel fluorescence palette (DAPI + 4 markers): path 2C with n>3 - Per-channel norm list: new feature, independent contrast per channel - Single-element norm list: verifies list[Normalize] works for 1 channel Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Reuse pre-computed zero_colors array instead of calling cmap(0) twice - Add tests for norm list length mismatch and empty norm list - Remove unconditional multi-cmap warning (signal-based blend handles it) - Clarify norms field naming with inline comment Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Replace stringly-typed `_na.alpha == "00"` with the public accessor `_na.get_alpha_as_float() == 0.0` in all three groups-filtering guards. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…ards - Reject tuple/sequence norms in _type_check_params with a clear TypeError pointing users to use a list instead - Extract na_is_transparent local in all three groups-filtering guards (shapes, points, labels) for consistent style - Add test_norm_tuple_raises Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Updates 5 existing baselines affected by the signal-based blending change and adds 4 new baselines for per-channel normalization and 5-channel compositing tests. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

LucaMarconato · 2026-03-19T19:23:26Z

Could this PR help with #534?

timtreis · 2026-03-19T19:26:01Z

Could this PR help with #534?

I think so 🤔

timtreis linked an issue Apr 22, 2025 that may be closed by this pull request

Feature: PCA aggregation for n >= 4 channels in render_images #450

Open

This comment was marked as outdated.

Sign in to view

timtreis marked this pull request as draft April 23, 2025 12:30

timtreis changed the title ~~Feature: PCA aggregation for n >= 4 channels in render_images~~ Fix multi-channel handling May 11, 2025

timtreis mentioned this pull request May 13, 2025

Clipping warning when plotting images #416

Closed

Sonja-Stockhaus pushed a commit that referenced this pull request Jun 4, 2025

code by Tim (#451), only the stacking approach

b739428

Sonja-Stockhaus mentioned this pull request Jun 4, 2025

fix image rendering (clipping warning) #471

Merged

timtreis force-pushed the bugfix/issue450-feature-pca-aggregation-for-n-4-channels-in-render_images branch from 2e991c5 to 92cb5eb Compare March 19, 2026 14:24

timtreis changed the title ~~Fix multi-channel handling~~ Fix multi-channel compositing and add per-channel norm support Mar 19, 2026

timtreis and others added 10 commits March 19, 2026 15:40

Update CI-generated baseline images for additive blending compositing

becae8d

Updates 5 baseline images that changed due to the switch from averaging to additive blending in multi-channel image compositing. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Use Color.get_alpha_as_float() instead of comparing internal hex string

fe75761

Replace stringly-typed `_na.alpha == "00"` with the public accessor `_na.get_alpha_as_float() == 0.0` in all three groups-filtering guards. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add CI-generated baseline images for signal-based compositing tests

09519d8

Updates 5 existing baselines affected by the signal-based blending change and adds 4 new baselines for per-channel normalization and 5-channel compositing tests. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Clean up comments: remove AI-style symbols and reword unclear comments

fd051a2

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add empty-layers guard to _additive_blend, clarify scalar_norm comment

11dbd79

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

timtreis mentioned this pull request Mar 19, 2026

Failing to recreate Xenium Exlorer-style multi-stain segmentation panel plot #534

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix multi-channel compositing and add per-channel norm support#451

Fix multi-channel compositing and add per-channel norm support#451
timtreis wants to merge 11 commits intomainfrom
bugfix/issue450-feature-pca-aggregation-for-n-4-channels-in-render_images

timtreis commented Apr 22, 2025 •

edited

Loading

Uh oh!

timtreis commented Apr 22, 2025

Uh oh!

This comment was marked as outdated.

Sonja-Stockhaus commented Apr 23, 2025

Uh oh!

codecov-commenter commented May 11, 2025 •

edited

Loading

Uh oh!

LucaMarconato commented Mar 19, 2026

Uh oh!

timtreis commented Mar 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

timtreis commented Apr 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Bug fixes

Per-channel norm usage

Test plan

Uh oh!

timtreis commented Apr 22, 2025

Uh oh!

This comment was marked as outdated.

Sonja-Stockhaus commented Apr 23, 2025

Uh oh!

codecov-commenter commented May 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

LucaMarconato commented Mar 19, 2026

Uh oh!

timtreis commented Mar 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

timtreis commented Apr 22, 2025 •

edited

Loading

codecov-commenter commented May 11, 2025 •

edited

Loading