feat(metrics): add binomial splitting for single-image FRC/FSC by alxndrkalinin · Pull Request #40 · alxndrkalinin/cubic

alxndrkalinin · 2026-04-01T05:21:00Z

Summary

Add binomial_split to image_utils.py: implements photon-count splitting (Rieger et al., Optics Express 2024) as an alternative to checkerboard for single-image FRC/FSC. Supports counts mode (with gain/offset/readout noise correction) and Poisson thinning fallback. Multi-repeat averaging via n_repeats gives correlation-std and resolution-std.
Thread split_type="binomial" through calculate_frc, frc_resolution, fsc_resolution, calculate_sectioned_fsc.
Fix code quality issues across the codebase: narrow exception handling, enforce mypy, fix README typos, proper type annotations without # type: ignore suppressions.

Test plan

pytest tests/metrics/spectral/test_frc.py — binomial split tests (223 lines)
pytest tests/test_image_utils.py — binomial_split unit tests (107 lines)
uv run mypy --ignore-missing-imports cubic/ — no new errors
uv run ruff check . && uv run ruff format --check . — clean

🤖 Generated with Claude Code

Summary by Sourcery

Add support for binomial photon-count-based splitting as an alternative to checkerboard for single-image FRC/FSC, including multi-repeat averaging with uncertainty estimates, while tightening typing, validation, and documentation across metrics, segmentation, and feature modules.

New Features:

Introduce binomial_split helper to split single images via binomial sampling in counts and Poisson-thinning modes, with optional camera calibration and readout-noise correction.
Extend FRC/FSC APIs to accept split_type, counts_mode, camera calibration parameters, RNG, and n_repeats, enabling binomial-based single-image correlation with correlation and resolution uncertainty outputs.
Expose per-ring correlation standard deviation and resolution standard deviation fields in FourierCorrelationData for multi-repeat analyses.

Bug Fixes:

Ensure spacing handling and shape usage are correct and explicit in DCR, FSC, and related spectral computations to avoid mismatches and ambiguous behavior.
Fix mesh feature extraction to handle missing or non-finite trimesh volume/area values more robustly.
Avoid mis-typed or ambiguous spacing and device conversions in voxel features, segmentation utilities, and deconvolution observers.

Enhancements:

Refactor FRC/FSC preprocessing and single-pass computation into clearer helpers, including deterministic RNG handling for repeatable binomial splits.
Tighten input validation and error handling in average-precision matching, segmentation utilities, and label operations by replacing asserts with explicit exceptions and trimming noisy prints.
Clarify type hints and internal APIs across CUDA management, segmentation, feature extraction, and spectral metrics for better mypy compliance.
Improve checkerboard-based resolution workflows (grid/five-crop, sectioned FSC) with clearer spacing requirements, padding logic, and cutoff correction control.

Documentation:

Document FRC/FSC image splitting strategies, including when to use checkerboard vs binomial splitting and how to configure counts vs Poisson-thinning modes, in the spectral metrics README.
Fix typos and wording in the main README and agent guidance files, and update CLAUDE/AGENTS documentation around contribution and commit practices.

Tests:

Add comprehensive unit tests for binomial_split, covering conservation, variance, independence, RNG determinism, and 2D/3D behavior.
Add FRC/FSC spectral tests to validate binomial splitting behavior, multi-repeat averaging, readout-noise correction, and compatibility with legacy checkerboard defaults.

Chores:

Enable stricter static typing and configuration (e.g., mypy) and clean up type-ignore usage and mutable globals where needed.

Add a procedural commit workflow checklist (list → group by concept → plan N commits → execute) to prevent grouping changes by workflow step. Simplify CLAUDE.md to just reference AGENTS.md. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Implement random binomial splitting (Rieger et al., Optics Express 2024) as an alternative to checkerboard splitting for single-image FRC/FSC. Each pixel's photon count is split into n1 ~ Binomial(n, 0.5) and n2 = n - n1, producing two noise-independent images with preserved Poisson statistics and no spatial subsampling. New `split_type="binomial"` parameter threads through calculate_frc, frc_resolution, fsc_resolution, and calculate_sectioned_fsc. Supports counts mode (raw camera data with gain/offset/readout noise correction) and poisson_thinning mode (fallback for float/deconvolved images). Multi-repeat averaging via `n_repeats` produces correlation-std and resolution-std for uncertainty quantification. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

- Use cubic.cuda abstraction (to_same_device/asnumpy) in binomial_split - Use nanmean/nanstd for multi-repeat FRC averaging - Add xy_std/z_std keys to single-pass binomial fsc_resolution - Document resolution averaging asymmetry in FRC - Add ndim, gain, readout_noise_rms validation to binomial_split - Fix gain convention docstring in binomial_split Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

- Warn when n_repeats>1 is silently ignored (checkerboard or two-image) - Clarify n_repeats docstring: requires single-image + binomial mode - Add xy_std/z_std keys to fsc_resolution mask backend return path Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

…, iterators.py - preprocess_images: assert image2 is not None in else branch and before hamming_window - iterators.py: accept int | float for d_extract_angle - calculate_sectioned_fsc: use eff_spacing for FourierCorrelationAnalysis, add shape3d for sectioned_bin_id - fsc_resolution: assert spacing_list is not None after _normalize_spacing guard - calculate_frc/fsc_resolution: assert original_image1 is not None in reverse/do_average blocks - grid_crop_resolution (both): raise ValueError for None spacing, use spacing_list for indexing, annotate aggregate_fn as Callable, rename pad_size to half/pad_arg to avoid int/tuple reassignment - image_utils.py: merge poisson_thinning img1/img2 assignments with to_same_device - bandlimited.py: type _APODIZATION_FNS as dict[str, Callable] to fix apo_fn operator errors Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

- Add targeted type-ignore comments for pre-existing mypy errors - Narrow exception handling in estimate_cutoff to (ValueError, RuntimeError, TypeError) - Replace print/assert with logging/ValueError in average_precision - Fix typos in README - Enforce mypy type checking (remove ignore_errors = true from pyproject.toml) - Add class-level annotations to CUDAManager, cast mesh volume - Align spacing to Sequence[float] in dcr.py internals - Replace ArrayLike with ndarray in segmentation modules Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

- Replace assert with RuntimeError in _calculate_frc_single_pass for original_image1 None check (assert stripped by python -O) - Warn when n_repeats>1 is silently ignored on deprecated mask backend in fsc_resolution Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

sourcery-ai · 2026-04-01T05:21:20Z

Reviewer's Guide

Adds a new binomial photon-count splitting method for single-image FRC/FSC (with counts and Poisson-thinning modes and multi-repeat uncertainty estimates), threads the new split configuration through FRC/FSC APIs, and tightens type-safety and error handling across metrics, segmentation, CUDA, and documentation to satisfy mypy/ruff constraints.

Class diagram for FourierCorrelationData with new statistics fields

classDiagram
    class FourierCorrelationData {
        +FixedDictionary correlation
        +FixedDictionary resolution
        +__init__(data)
    }

    class FourierCorrelationDataCollection {
        +__getitem__(index) FourierCorrelationData
        +__setitem__(index, value) void
        +__iter__() FourierCorrelationData
    }

    class FixedDictionary {
        -keys: list~str~
        -data: dict~str, Any~
        +__getitem__(key) Any
        +__setitem__(key, value) void
    }

    FourierCorrelationData --> FixedDictionary : uses correlation
    FourierCorrelationData --> FixedDictionary : uses resolution

    %% Correlation keys
    class CorrelationFields {
        +correlation: np.ndarray
        +frequency: np.ndarray
        +points_x_bin: np.ndarray
        +curve_fit: np.ndarray
        +curve_fit_coefficients: np.ndarray
        +correlation_std: np.ndarray
    }

    %% Resolution keys
    class ResolutionFields {
        +threshold: float
        +criterion: str
        +resolution_point: float
        +resolution_threshold_coefficients: np.ndarray
        +resolution: float
        +spacing: float
        +resolution_std: float
    }

    FourierCorrelationData o-- CorrelationFields : logical_fields
    FourierCorrelationData o-- ResolutionFields : logical_fields

File-Level Changes

Change	Details	Files
Introduce binomial photon-count splitting utility and wire it into FRC/FSC preprocessing and analysis with multi-repeat support and uncertainty outputs.	Add binomial_split image utility supporting counts and poisson_thinning modes, camera gain/offset/readout-noise correction, and RNG control, with numpy/cupy device handling. Extend FRC/FSC preprocessing (preprocess_images) to choose between checkerboard and binomial splitting for single-image workflows, propagating counts_mode, gain, offset, readout_noise_rms, and rng parameters. Refactor FRC computation into a reusable _calculate_frc_single_pass helper that handles spacing normalization, checkerboard reverse-split averaging, and is reused for both single-pass and repeated binomial runs. Update calculate_frc, frc_resolution, calculate_sectioned_fsc, and fsc_resolution APIs to accept split_type, counts_mode, camera calibration, n_repeats, and rng, implementing binomial multi-repeat averaging (with correlation-std and resolution-std) and logging/warnings when options are ignored. Adjust FSC histogram backend helpers and resolution extraction (_fsc_hist_compute, _fsc_extract_resolution) to respect split_type when deciding on checkerboard cutoff correction and averaging, and to support binomial repeats in the hist backend.	`cubic/image_utils.py` `cubic/metrics/spectral/frc.py`
Expand FRC/FSC tests to cover binomial splitting behavior, statistics, and backwards compatibility, plus unit tests for the new binomial_split utility.	Add synthetic Poisson image generators and FRC tests verifying binomial basic validity, lack of checkerboard cutoff correction, n_repeats behavior, statistical equivalence of 1FRC vs 2FRC, and readout-noise plateau correction. Add FSC tests for 3D binomial splitting and ensure checkerboard remains the default/unchanged when split_type is omitted. Introduce a dedicated TestBinomialSplit test class exercising conservation, expectation, variance scaling, independence in poisson_thinning, determinism with seeds, non-negativity with readout correction, and 3D handling, plus input validation and warning behavior.	`tests/metrics/spectral/test_frc.py` `tests/test_image_utils.py`
Improve spectral metrics and DCR internals to be more type-safe and explicit about spacing, iterator parameters, and data structures.	Extend FourierCorrelationData fixed dictionaries with correlation-std and resolution-std keys to carry uncertainty from binomial repeats. Normalize and validate spacing handling in FRC/FSC and DCR helpers, including explicit spacing_list variables, Sequence[float] annotations, and proper passing into radial_edges and sectioned_bin_id. Relax HollowSectionedFourierShellIterator d_extract_angle type to accept both int and float for improved typing flexibility. Fix sectioned FSC histogram implementation to always pass a 3D shape into sectioned_bin_id and to construct effective spacing when None is given.	`cubic/metrics/spectral/analysis.py` `cubic/metrics/spectral/frc.py` `cubic/metrics/spectral/dcr.py` `cubic/metrics/spectral/iterators.py`
Tighten error handling and typing across metrics and segmentation utilities to satisfy mypy and avoid silent failures.	Refactor average_precision label-overlap implementation to hoist numba JIT outside the function, add a module-level logger, avoid printing, and raise ValueError instead of asserts for invalid inputs or missing matches. Change various segmentation utilities (downscale_and_filter, cleanup_segmentation, remove_large_objects/small_objects, clear_xy_borders, remove_touching_objects, segment_watershed, fill_holes_slicer) and cellpose wrappers to use concrete np.ndarray typing and clarify return types. Add runtime type assertions where necessary (e.g., distance_transform_edt outputs, iter_image in deconvolution observer) and initialize typed attributes in CUDAManager; annotate apodization function registry with concrete Callable type. Replace unsafe asserts with proper exceptions in matching/average-precision code and watershed helpers to avoid assertion-stripping behavior in optimized runs.	`cubic/metrics/average_precision.py` `cubic/segmentation/segment_utils.py` `cubic/segmentation/cellpose.py` `cubic/preprocessing/deconvolution.py` `cubic/metrics/bandlimited.py` `cubic/cuda.py`
Update documentation and project configuration to describe splitting methods, commit workflow, and stricter static checks.	Extend spectral metrics README with a new Image Splitting Methods section comparing checkerboard vs binomial, usage examples for FRC/FSC, and guidance on counts vs poisson_thinning modes with literature reference. Clarify top-level README wording and typos around GPU acceleration, Fourier Ring/Shell correlation, and average precision features. Revise AGENTS.md and CLAUDE.md to instruct AI agents to read AGENTS and follow a structured commit workflow based on conceptual grouping rather than workflow steps. Tighten DCR public API typing and return annotations, and mark trimesh import fallback in mesh features with type: ignore to appease mypy; fix mesh feature computations to cast potentially None-like volumes/areas to float and reuse scalar volume in derived metrics. Remove mypy ignore_errors flag from pyproject.toml to enforce static type checking, and adjust voxel regionprops helpers to avoid in-place spacing mutation and pass tuple spacing explicitly for cucim compatibility.	`cubic/metrics/spectral/README.md` `README.md` `AGENTS.md` `CLAUDE.md` `cubic/feature/mesh.py` `cubic/feature/voxel.py` `cubic/metrics/spectral/dcr.py` `pyproject.toml`

Tips and commands

Interacting with Sourcery

Trigger a new review: Comment @sourcery-ai review on the pull request.
Continue discussions: Reply directly to Sourcery's review comments.
Generate a GitHub issue from a review comment: Ask Sourcery to create an
issue from a review comment by replying to it. You can also reply to a
review comment with @sourcery-ai issue to create an issue from it.
Generate a pull request title: Write @sourcery-ai anywhere in the pull
request title to generate a title at any time. You can also comment
@sourcery-ai title on the pull request to (re-)generate the title at any time.
Generate a pull request summary: Write @sourcery-ai summary anywhere in
the pull request body to generate a PR summary at any time exactly where you
want it. You can also comment @sourcery-ai summary on the pull request to
(re-)generate the summary at any time.
Generate reviewer's guide: Comment @sourcery-ai guide on the pull
request to (re-)generate the reviewer's guide at any time.
Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
pull request to resolve all Sourcery comments. Useful if you've already
addressed all the comments and don't want to see them anymore.
Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
request to dismiss all existing Sourcery reviews. Especially useful if you
want to start fresh with a new review - don't forget to comment
@sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

Enable or disable review features such as the Sourcery-generated pull request
summary, the reviewer's guide, and others.
Change the review language.
Add, remove or edit custom review instructions.
Adjust other review settings.

Getting Help

Contact our support team for questions or feedback.
Visit our documentation for detailed guides and information.
Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

sourcery-ai

Hey - I've found 3 issues, and left some high level feedback:

The informational log on n_repeats == 1 for binomial splits (_BINOMIAL_SINGLE_REPEAT_MSG) is emitted on every call and may be noisy in real pipelines; consider downgrading this to logger.debug, or gating it with a one-time flag or using warnings.warn if you specifically want user-facing feedback.
The binomial FRC/FSC parameter set (split_type, counts_mode, gain, offset, readout_noise_rms, n_repeats, rng) is threaded manually through several public functions; consider introducing a small configuration object or helper to reduce duplication and the risk of parameter drift across these call sites.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- The informational log on `n_repeats == 1` for binomial splits (`_BINOMIAL_SINGLE_REPEAT_MSG`) is emitted on every call and may be noisy in real pipelines; consider downgrading this to `logger.debug`, or gating it with a one-time flag or using `warnings.warn` if you specifically want user-facing feedback.
- The binomial FRC/FSC parameter set (`split_type`, `counts_mode`, `gain`, `offset`, `readout_noise_rms`, `n_repeats`, `rng`) is threaded manually through several public functions; consider introducing a small configuration object or helper to reduce duplication and the risk of parameter drift across these call sites.

## Individual Comments

### Comment 1
<location path="cubic/preprocessing/deconvolution.py" line_range="301-303" />
<code_context>

     for i, result in enumerate(results):
-        results[i]["iter_image"] = asnumpy(result["iter_image"])
+        iter_image = result["iter_image"]
+        assert isinstance(iter_image, np.ndarray)
+        results[i]["iter_image"] = asnumpy(iter_image)

     return (thresh_iter, results)
</code_context>
<issue_to_address>
**issue (bug_risk):** The new ndarray assertion in `decon_observer` may break GPU / arraylike results.

This assertion changes behavior: `result['iter_image']` used to be passed directly to `asnumpy`, allowing any arraylike your CUDA helpers support (e.g., CuPy). Now `assert isinstance(iter_image, np.ndarray)` will fail for CuPy or other non-NumPy arrays that `asnumpy` can still handle. If you just want to guard against `None` or non-array values, consider dropping the assertion, checking minimal properties (e.g., shape/dtype), or using a duck-typed check (e.g., `hasattr(iter_image, 'shape')`) instead of requiring `np.ndarray`.
</issue_to_address>

### Comment 2
<location path="tests/metrics/spectral/test_frc.py" line_range="524-533" />
<code_context>
+def test_frc_binomial_readout_noise_plateau() -> None:
</code_context>
<issue_to_address>
**issue (testing):** Readout noise plateau test assertion does not match the stated expectation

The docstring states that "Readout noise correction reduces high-frequency FRC plateau bias", but `tail_with <= tail_no + 0.05` permits the corrected plateau to be slightly higher than the uncorrected one. To better enforce the intended behavior, consider an assertion like `tail_with <= tail_no` (optionally with a small tolerance for numerical noise, e.g. `tail_with <= tail_no * 1.01`), so the test fails when the correction degrades the plateau instead of allowing non-improving or slightly worse outcomes to pass.
</issue_to_address>

### Comment 3
<location path="CLAUDE.md" line_range="3" />
<code_context>
-This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
-
[email protected]
+Read @AGENTS.md file and follow instructions closely.
</code_context>
<issue_to_address>
**nitpick (typo):** Add missing articles/possessive for smoother grammar in this sentence.

Consider rephrasing to: `Read the @AGENTS.md file and follow its instructions closely.` This fixes the missing article and possessive while keeping the same meaning.
</issue_to_address>

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

- grid_crop_resolution / five_crop_resolution: replace assert for input validation with raise ValueError (asserts are stripped by python -O) - fsc_resolution: replace assert spacing_list is not None with RuntimeError - Remove narration comment "No cutoff correction for binomial" in calculate_frc Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

- Extract calibration warning thresholds as named constants in image_utils.py - Add clipping-fraction warning after readout noise correction — large readout_noise_rms relative to photon counts zeros many pixels silently - Replace assert image2 is not None with raise RuntimeError in preprocess_images — asserts are stripped by python -O - Warn when counts_mode != 'counts' with split_type='checkerboard' in calculate_frc and frc_resolution — params are silently ignored there - Document RNG state mutation in _make_repeat_rngs docstring Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

…s_mode warning - test_frc_binomial_camera_calibration: verifies gain/offset/readout_noise_rms flow through frc_resolution end-to-end with simulated raw camera data - test_counts_mode_warns_without_binomial: verifies UserWarning is raised when counts_mode='poisson_thinning' is passed with split_type='checkerboard' Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

n1_cpu is int64; comparing against float32 half_var upcasts to float64, giving a numerically different result than what actually gets clipped (float32 img1). Cast to float32 before comparison to match the clipping dtype. Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

assert isinstance(iter_image, np.ndarray) fails for CuPy arrays, which asnumpy() handles correctly. The assert was overly strict and violated the device-agnostic contract of the module. Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

The fixed seed produces tail_with=0.039 vs tail_no=0.121, so the +0.05 additive slack was masking potential regressions where the correction makes things worse. Remove the slack to enforce that corrected plateau is strictly no worse than uncorrected. Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

Copilot

Pull request overview

Adds a binomial photon-count splitting strategy for single-image FRC/FSC (as an alternative to checkerboard), threads the new parameters through spectral metrics APIs, and tightens typing/docs/tests across the project.

Changes:

Add binomial_split and integrate split_type="binomial" into FRC/FSC workflows (incl. multi-repeat uncertainty outputs).
Extend/adjust spectral metrics (FRC/FSC/DCR) typing, spacing handling, and internal helpers.
Add/expand tests and documentation; tighten mypy configuration and various type hints.

Reviewed changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tests/test_image_utils.py	Adds unit tests for `binomial_split` behavior and edge cases.
tests/metrics/spectral/test_frc.py	Adds FRC/FSC tests covering binomial splitting, repeats, calibration/read-noise behavior.
README.md	Fixes typos/wording in project overview.
pyproject.toml	Enables stricter mypy by removing `ignore_errors`.
cubic/segmentation/segment_utils.py	Updates type hints and watershed flow; adds assertions and refactors CPU masking.
cubic/segmentation/cellpose.py	Replaces `npt.ArrayLike` with `np.ndarray` typing in signatures.
cubic/metrics/spectral/README.md	Documents splitting methods and adds binomial usage guidance.
cubic/metrics/spectral/iterators.py	Broadens `d_extract_angle` type to allow floats.
cubic/metrics/spectral/frc.py	Implements binomial split plumbing, repeat averaging, and cutoff behavior changes.
cubic/metrics/spectral/dcr.py	Adjusts spacing typing and sectioned DCR internals.
cubic/metrics/spectral/analysis.py	Extends result containers to include std fields.
cubic/metrics/bandlimited.py	Tightens typing for apodization function registry.
cubic/metrics/average_precision.py	Hoists optional numba JIT helper and replaces asserts with exceptions.
cubic/image_utils.py	Introduces `binomial_split` implementation and related calibration warnings.
cubic/feature/voxel.py	Clarifies spacing handling for cucim memoization.
cubic/feature/mesh.py	Attempts to harden mesh feature extraction against missing volume values.
cubic/cuda.py	Adds annotations for `CUDAManager` singleton fields.
CLAUDE.md	Updates contributor guidance pointer to AGENTS.md.
AGENTS.md	Adds a commit workflow checklist.

Comments suppressed due to low confidence (1)

cubic/metrics/spectral/dcr.py:530

This adds a new # type: ignore[arg-type] when calling sectioned_bin_id. Since sectioned_bin_id requires a tuple[int, int, int], consider constructing a properly-typed shape3d = (int(shape[0]), int(shape[1]), int(shape[2])) (similar to the approach used in frc.py) so mypy passes without suppressions.

    # Get sectioned bin IDs
    radial_id, angle_id = sectioned_bin_id(
        shape,  # type: ignore[arg-type]
        r_edges,
        angle_edges,
        spacing=list(spacing) if spacing is not None else None,
        exclude_axis_angle=exclude_axis_angle,

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Resolve conflicts from 8883d9c landing on main after our branch diverged. All conflicts resolved by keeping HEAD (superset of main's changes) plus applying main's label() -> np.ndarray return type fix that was lost in the conflict block. Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

…nt_watershed Same pattern as deconvolution.py: assert isinstance(distance, np.ndarray) fails for CuPy arrays returned by distance_transform_edt on GPU. Replace with targeted type: ignore comments on the downstream usages that mypy cannot narrow without the assert. Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

- frc.py: fix gain docstring in calculate_frc and fsc_resolution — gain is ADU/electron (electrons = (image - offset) / gain), not electrons/ADU as previously stated - mesh.py: replace `float(mesh.volume or 0.0)` with an explicit isfinite guard; NaN is truthy in Python so `nan or 0.0` == nan, which propagates silently through Extent/Solidity divisions Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

When n_repeats > 1, the averaged mean FRC curve can stay above the threshold at all frequencies (especially with poisson_thinning on moderate-signal data), yielding NaN for the mean-curve resolution even though individual repeats produce valid resolutions. Fall back to the mean of per-repeat resolutions, matching the FSC multi-repeat path which already uses nanmean. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Side-by-side comparison of the two single-image splitting strategies for FRC (2D) and FSC (3D): checkerboard (Koho 2019) vs binomial (Rieger 2024). Demonstrates poisson_thinning mode, multi-repeat uncertainty quantification, and directional 3D FSC on pollen data. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

…on floor Binomial counts split produces anticorrelated noise (n1 + n2 = n). The FRC hist backend took abs() of the cross-spectrum numerator, flipping negative noise correlation to a false ~0.3 positive floor at high frequencies. Preserve the sign when signed=True so the FRC correctly goes negative in the noise-dominated regime. The fix threads a signed flag from calculate_frc through _calculate_frc_single_pass to _calculate_frc_core. It is enabled only for split_type="binomial" + counts_mode="counts". Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

…odes - Show all three FRC variants: checkerboard, binomial counts, binomial poisson_thinning (was only showing poisson_thinning) - Fix bug: res_checker was assigned from frc_binom instead of frc_checker - Explain why poisson_thinning gives different results (measures noise model self-consistency, not physical resolution) - Note that checkerboard matches Koho et al. 2019 published values - Set y-axis to [-0.5, 1.05] to show signed FRC negative tail Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Add low-frequency bin exclusion for DC / background / autofluorescence suppression. nbins_low zeros the first N bins (hard cutoff); taper_low applies a half-cosine ramp (soft transition). taper_low takes precedence when both are set. Ported from feat/experimental-spectral-metrics without the experimental weighting methods (smooth_wiener, snr2, etc.) that remain parked. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Copilot

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

The half-cosine ramp used arange(1, n+1)/n which started at 1/n instead of 0, so taper_low=1 had multiplier 1.0 (no taper) and DC was never fully suppressed. Use arange(n)/(n-1) so the first bin is 0 and the last bin is 1, with a special case for n=1. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

alxndrkalinin and others added 7 commits March 31, 2026 16:54

sourcery-ai Bot reviewed Apr 1, 2026

View reviewed changes

Comment thread cubic/preprocessing/deconvolution.py

Comment thread tests/metrics/spectral/test_frc.py

Comment thread CLAUDE.md Outdated

alxndrkalinin and others added 7 commits March 31, 2026 22:39

docs: fix grammar in CLAUDE.md

c6a3fa0

Co-Authored-By: Claude Sonnet 4.6 (1M context) <[email protected]>

alxndrkalinin requested a review from Copilot April 2, 2026 00:38

Copilot started reviewing on behalf of alxndrkalinin April 2, 2026 00:39 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Comment thread cubic/segmentation/segment_utils.py Outdated

Comment thread cubic/metrics/spectral/frc.py Outdated

Comment thread cubic/metrics/spectral/frc.py Outdated

Comment thread cubic/feature/mesh.py

Comment thread cubic/image_utils.py

alxndrkalinin and others added 8 commits April 1, 2026 17:44

alxndrkalinin requested a review from Copilot April 2, 2026 21:40

Copilot started reviewing on behalf of alxndrkalinin April 2, 2026 21:41 View session

docs: add split comparison notebook to README examples table

9f71604

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Comment thread cubic/metrics/bandlimited.py Outdated

alxndrkalinin merged commit 207b6bd into main Apr 2, 2026
9 checks passed

alxndrkalinin deleted the feat/frc-binomial-split branch April 2, 2026 21:54

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(metrics): add binomial splitting for single-image FRC/FSC#40

feat(metrics): add binomial splitting for single-image FRC/FSC#40
alxndrkalinin merged 24 commits intomainfrom
feat/frc-binomial-split

alxndrkalinin commented Apr 1, 2026 •

edited by sourcery-ai Bot

Loading

Uh oh!

sourcery-ai Bot commented Apr 1, 2026 •

edited

Loading

Interacting with Sourcery

Customizing Your Experience

Getting Help

Uh oh!

sourcery-ai Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

alxndrkalinin commented Apr 1, 2026 • edited by sourcery-ai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Summary by Sourcery

Uh oh!

sourcery-ai Bot commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviewer's Guide

Class diagram for FourierCorrelationData with new statistics fields

File-Level Changes

Interacting with Sourcery

Customizing Your Experience

Getting Help

Uh oh!

sourcery-ai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

alxndrkalinin commented Apr 1, 2026 •

edited by sourcery-ai Bot

Loading

sourcery-ai Bot commented Apr 1, 2026 •

edited

Loading