Add evaluation notebook and metrics implementation for degraded image…

[wli51]

Adding notebook for actual metric eval with metrics from #7. Also adds and evaluates two additional foreground metric variants.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[Copilot]

Pull request overview

Adds an analysis notebook to evaluate full-reference IQA metrics on simulated degradation/reference Lance pairs, and extends the metric suite with foreground-masked SSIM/PSNR variants (Otsu mask from the reference) to better focus evaluation on salient pixels in microscopy images.

Changes:

• Implement ForegroundSSIM and ForegroundPSNR in utils/custom_metrics.py using a per-image Otsu-derived foreground mask.
• Add notebook + nbconverted script to run metric evaluation over paired Lance datasets and write results to Parquet.
• Update READMEs to document the metric set and the new foreground-masked variants; add kornia to dependencies.

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
utils/custom_metrics.py	Adds foreground-masked SSIM/PSNR metric implementations built around fixed-range per-image Otsu thresholding.
pyproject.toml	Adds kornia to project dependencies.
2.train_models/README.md	New README describing goals/approach/metrics for the analysis workflow.
1.image_degradation_simulation/README.md	Documents foreground-masked metric behavior (Otsu mask, NaN on empty foreground).
1.image_degradation_simulation/nbconverted/1.4.evaluate_metrics_on_degraded_image.py	Adds a script version of the evaluation notebook for metric computation and Parquet output.
1.image_degradation_simulation/1.4.evaluate_metrics_on_degraded_image.ipynb	Adds the evaluation notebook to run the metrics over paired reference/degraded Lance datasets.

Suppressed comments (3)

2.train_models/README.md:27

• Grammar/pluralization: “36 degraded variant per image” should be plural.

This produces 36 degraded variant per image in our Cell Painting dataset.

2.train_models/README.md:44

• Spelling/grammar: “Per every… varaint” contains typos and is awkwardly phrased.

Per every pair of degraded image variant and its original reference image, we evaluate our collection of 7 metrics, yielding 7 metric values per degraded varaint.

2.train_models/README.md:50

• Multiple spelling errors in the variance-partitioning description (“variantions”, “indenitity”, “densitiy”).

We then analyze with ANOVA variance partitioning and nested regression to what extents are variantions in metric values explained by the known degradation severity changes,
cell line indenitity of the images,
seeding densitiy label associated with the images,
and other known data inherent factors.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.