Prior inspect peaks work by mgiammar · Pull Request #124 · Lucaslab-Berkeley/Leopard-EM

mgiammar · 2026-04-24T22:42:47Z

No description provided.

jdickerson95 · 2026-04-30T20:46:53Z

@mgiammar I've implemented the inspect peaks, including frc and frame correlations.
I've also updated Leopard-EM to use the new deformation field class, so the tests will definitely fail until a new version of the monorepo is released.

…uted match-template Squashed rebuild of the mdg_inspect_peaks PR (Lucaslab-Berkeley#124) on top of development_v1.3 to remove duplicated upstream history. Tree is identical to prior branch tip 84d1ce8.

mgiammar · 2026-06-08T22:03:30Z

@jdickerson95 have updated the TeamTomo dependencies to >=v0.5.2 so those aspects of CI should pass now

mgiammar · 2026-06-08T22:48:08Z

+from torch_motion_correction.deformation_field import (  # pyright: ignore[reportMissingImports]
+    DeformationField,
+)


Import now exists

mgiammar · 2026-06-08T22:50:07Z

+            s=self.extracted_box_size,
+            dim=(-2, -1),
+        )
+        return cast(torch.Tensor, particle_movie.permute(1, 0, 2, 3).contiguous())


Is cast call here necessary?

…lated components

Copilot

Pull request overview

This PR introduces a peak-inspection workflow that exposes full local scoring tensors (cross-correlation maps or FRC spectra) for offline inspection, and adds a per-frame inspection manager for movie frame-by-frame scoring with optional fixed whitening/normalization reuse. It also refactors parts of the refine/inspect backend setup for multi-device execution and adds random Fourier dropout support in preprocessing.

Changes:

Add PeakInspectionManager / FrameInspectionManager plus CLI examples and config templates for inspecting local score tensors and per-frame results.
Extend backend correlation codepaths to optionally skip projection normalization and to compute FRC outputs.
Refactor image filtering/normalization utilities and backend setup helpers; add tests to lock in reducer/stacking behavior.

Reviewed changes

Copilot reviewed 29 out of 29 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/pydantic_models/test_particle_stack.py	Adds coverage for aligned-movie particle extraction and movie→image stack API behavior.
tests/pydantic_models/test_frame_inspection_manager.py	New unit tests for frame inspection reduction/stacking and CSV output behavior.
tests/backend/test_core_refine_template.py	New deterministic snapshot test for `_reduce_refine_best_zscore`.
tests/backend/test_core_match_template.py	Avoids division-by-zero in set-overlap assertions.
src/leopard_em/utils/image_processing.py	Splits normalization into a reusable `get_image_normalization_factor` and adds optional precomputed filter/normalization inputs.
src/leopard_em/utils/ctf_utils.py	Adds helper to move tensor-valued CTF kwargs onto a target device.
src/leopard_em/utils/backend_setup.py	Refactors backend input setup; adds per-frame filter setup helper and deformation-field type updates.
src/leopard_em/pydantic_models/results/match_template_result.py	Tweaks error message formatting for write-permission / overwrite checks.
src/leopard_em/pydantic_models/managers/refine_template_manager.py	Allows optional `template_tensor` override and updates deformation-field handling.
src/leopard_em/pydantic_models/managers/peak_inspection_manager.py	New manager to run inspect backend and optionally save self-describing outputs.
src/leopard_em/pydantic_models/managers/match_template_manager.py	Disables random Fourier dropout for image-side filtering during match-template setup.
src/leopard_em/pydantic_models/managers/frame_inspection_manager.py	New per-frame inspection manager (frame iteration, fixed filters, CSV summaries, optional dose-weighting).
src/leopard_em/pydantic_models/managers/init.py	Exports the new managers.
src/leopard_em/pydantic_models/data_structures/particle_stack.py	Adds per-frame particle movie extraction; updates deformation-field integration and refactors movie extraction internals.
src/leopard_em/pydantic_models/config/movie_config.py	Switches deformation field loading to `DeformationField.from_csv`.
src/leopard_em/pydantic_models/config/correlation_filters.py	Adds `RandomFourierDropoutConfig` and integrates it into combined filters.
src/leopard_em/backend/cross_correlation.py	Adds optional projection normalization toggle and implements batched FRC scoring.
src/leopard_em/backend/core_refine_template.py	Refactors multi-device tensor movement/stream handling; factors correlation batching and reducer logic.
src/leopard_em/backend/core_differentiable_refine.py	Ensures tensor-valued CTF kwargs are moved onto the active device before CTF stack construction.
src/leopard_em/analysis/inspect_peaks.py	New inspect backend entrypoint producing full local hypothesis tensors (CC or FRC), multi-device compatible.
src/leopard_em/analysis/inspect_peaks_result.py	New save/load helpers for self-describing `.npz` inspection outputs.
src/leopard_em/analysis/init.py	Re-exports inspection result save/load helpers.
pyproject.toml	Bumps required Python to 3.11 and updates TeamTomo dependency versions / adds FRC dependency.
programs/inspect_peaks/run_inspect_peaks.py	New example script for peak inspection producing a self-describing `.npz`.
programs/inspect_peaks/run_frame_inspection.py	New example script for per-frame inspection producing refine-like CSVs.
programs/inspect_peaks/inspect_peaks_example_config.yaml	New example YAML config for peak inspection.
programs/inspect_peaks/frame_inspection_example_config.yaml	New example YAML config for per-frame inspection.
.github/workflows/python-publish.yml	Updates CI python versions and bumps GitHub release action major version.
.github/workflows/pylint.yml	Updates lint workflow python versions.

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

+        do_random_masking = self.preprocessing_filters.random_fourier_dropout.enabled
+        self.preprocessing_filters.random_fourier_dropout.enabled = False
        cumulative_filter_image = self.preprocessing_filters.get_combined_filter(
            ref_img_rfft=image_dft,
            output_shape=image_dft.shape,
        )
+        self.preprocessing_filters.random_fourier_dropout.enabled = do_random_masking


mgiammar · 2026-06-15T15:18:31Z

Per-frame peak inspection should be updated in concert with the spatial peak inspection program to emit a npz file. May also need to address the underlying manager behavior.

mgiammar · 2026-06-15T15:20:51Z

How does frame inspection play into this results file? Could have a (num_particles, num_frames, ...) array rather than just a (num_particles, ...) array.

mgiammar · 2026-06-15T15:25:10Z

+    base_euler_angles : torch.Tensor | None, optional
+        Per-particle base ZYZ angles the offsets are relative to, shape ``(N, 3)``.


Should be a required argument, plus need to add particle defocus as a parameter

mgiammar · 2026-06-15T15:27:13Z

+        df = self.particle_stack._df  # pylint: disable=protected-access
+        particle_index = (
+            df["particle_index"].to_numpy() if "particle_index" in df.columns else None
+        )


NOTE: This will change with the other large backend refactor PR on dev-1.3 to prevent protected access

jdickerson95 marked this pull request as ready for review April 30, 2026 20:44

mgiammar added the enhancement New feature or request label May 16, 2026

feat: inspect peaks, fourier dropout, frame correlations, and distrib…

5e97fce

…uted match-template Squashed rebuild of the mdg_inspect_peaks PR (Lucaslab-Berkeley#124) on top of development_v1.3 to remove duplicated upstream history. Tree is identical to prior branch tip 84d1ce8.

mgiammar force-pushed the mdg_inspect_peaks branch from 84d1ce8 to 5e97fce Compare June 8, 2026 22:02

mgiammar commented Jun 8, 2026

View reviewed changes

mgiammar added 4 commits June 9, 2026 11:37

feat: pull out mean/std normalization into z-score from refinement core

b710cfb

feat: implement peak inspection result saving/loading and refactor re…

61291d4

…lated components

refactor: streamline peak inspection methods

5e61241

refactor: major restructuring of the per-frame inspect peaks manager

17335e3

mgiammar requested a review from Copilot June 15, 2026 14:57

Copilot started reviewing on behalf of mgiammar June 15, 2026 14:58 View session

Copilot AI reviewed Jun 15, 2026

View reviewed changes

mgiammar commented Jun 15, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Prior inspect peaks work#124

Prior inspect peaks work#124
mgiammar wants to merge 5 commits into
Lucaslab-Berkeley:development_v1.3from
mgiammar:mdg_inspect_peaks

mgiammar commented Apr 24, 2026

Uh oh!

jdickerson95 commented Apr 30, 2026

Uh oh!

mgiammar commented Jun 8, 2026

Uh oh!

mgiammar Jun 8, 2026

Uh oh!

mgiammar Jun 8, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

mgiammar Jun 15, 2026

Uh oh!

mgiammar Jun 15, 2026

Uh oh!

mgiammar Jun 15, 2026

Uh oh!

mgiammar Jun 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		base_euler_angles : torch.Tensor \| None, optional
		Per-particle base ZYZ angles the offsets are relative to, shape ``(N, 3)``.

Uh oh!

Conversation

mgiammar commented Apr 24, 2026

Uh oh!

jdickerson95 commented Apr 30, 2026

Uh oh!

mgiammar commented Jun 8, 2026

Uh oh!

mgiammar Jun 8, 2026

Choose a reason for hiding this comment

Uh oh!

mgiammar Jun 8, 2026

Choose a reason for hiding this comment

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

Reviewed changes

Uh oh!

mgiammar Jun 15, 2026

Choose a reason for hiding this comment

Uh oh!

mgiammar Jun 15, 2026

Choose a reason for hiding this comment

Uh oh!

mgiammar Jun 15, 2026

Choose a reason for hiding this comment

Uh oh!

mgiammar Jun 15, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants