Fix prediction mask names and background value

[vojtech-cifka]

Summary

• preserve original slide basenames for prediction TIFF maps
• remove generated hash suffixes from prediction mask filenames
• keep Unicode characters in slide names, e.g. MUG German slide names
• set prediction-map background pixels to 0
• make prediction-map background value explicit in LBFGS prediction-map configs

Motivation

Prediction masks were difficult to map back to their source WSIs because filenames had generated hash suffixes and non-ASCII characters were
sanitized. This broke names such as 61 Follikuläres Schilddrüsenkarzinom.

Prediction masks also used 255 for uncovered/background pixels, while the report-facing annotation masks use 0 after remapping. This caused
report overlays to render the whole non-predicted slide area as white.

Summary by CodeRabbit

• Chores
  • Updated default background value for prediction map generation from 255 to 0.
  • Modified prediction map output filename generation to derive names directly from input file paths rather than content-based naming.

[coderabbitai]

📝 Walkthrough

Walkthrough

TiffPredictionMapWriter changes default background value from 255 to 0, removes blake2b-based filename generation for simpler path-based naming, and configuration files explicitly set the background value to match the new implementation default.

Changes

TIFF Prediction Map Writer Update

Layer / File(s)	Summary
TiffPredictionMapWriter implementation updates ml/callbacks/tiff_prediction_map_writer.py	TiffPredictionMapWriter constructor default for background_value changes from 255 to 0, the blake2b import is removed, and _slide_prediction_filename is refactored to derive output TIFF names directly from input paths instead of generating digest-based filenames.
Configuration updates for background value configs/ml/trainer/final_with_prediction_maps.yaml, configs/experiment/ml/test_linear_virchow2_lbfgs.yaml	Both trainer and experiment configuration files explicitly add background_value: 0 to the tiff_prediction_maps callback configuration.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐰 A rabbit writes with care today,
Backgrounds now default to gray,
Blake's digest gone, filenames flow,
From simple paths, not hashes bright—
Configuration and code unite! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 33.33% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely summarizes the main changes: fixing prediction mask names (removing hash suffixes) and changing the background value from 255 to 0.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/mug-prediction-masks

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@ml/callbacks/tiff_prediction_map_writer.py`:
- Around line 519-520: _current implementation of _slide_prediction_filename
returns only the basename which causes silent overwrites for identically-named
files from different directories; update _slide_prediction_filename to preserve
the readable basename but append a stable, short disambiguator derived from the
full input path (e.g., first 8 chars of a hash of str(Path(path).resolve()) or
include parent folder name) before the .tiff suffix so names remain
deterministic and human-readable, and/or add a duplicate-detection check in the
writer that raises an error if two inputs would map to the same output; ensure
you reference and change the _slide_prediction_filename function and the caller
that writes prediction files so the new name format is used consistently.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: f2437f6d-25d0-430a-ab4e-c58d20390ed7

📥 Commits

Reviewing files that changed from the base of the PR and between 8a8f76e and ab1c90a.

📒 Files selected for processing (3)

• configs/experiment/ml/test_linear_virchow2_lbfgs.yaml
• configs/ml/trainer/final_with_prediction_maps.yaml
• ml/callbacks/tiff_prediction_map_writer.py

[gemini-code-assist]

Code Review

This pull request updates the TiffPredictionMapWriter by changing the default background_value from 255 to 0 in both the configuration files and the class constructor. Additionally, it simplifies the _slide_prediction_filename function by removing the hash suffix and previous sanitization logic. Feedback suggests that this simplification could lead to filename collisions and filesystem compatibility issues, recommending a more robust sanitization approach that handles illegal characters while maintaining Unicode support.