feat(drift): rigid drift correction — first feature mined from the quarry - #130
Draft
CSSFrancis wants to merge 6 commits into
Draft
feat(drift): rigid drift correction — first feature mined from the quarry#130CSSFrancis wants to merge 6 commits into
CSSFrancis wants to merge 6 commits into
Conversation
Pulled from feat/seg-fast-engine (PR #122) at file level; these are the behavioral spec for the rigid drift PR. test_drift_wizard.py is pruned of its non-rigid coverage (three TestMethodStubs tests: nonrigid selectable, nonrigid field parameterisation, unknown-field fallback) -- non-rigid lands in its own later PR. test_drift_translation.py is rigid-only and comes over verbatim.
…izard Pulled file-level from feat/seg-fast-engine (PR #122) and trimmed to the rigid surface: - spyde/drift/{frames,translation,warp}.py verbatim (all rigid): streaming frame source, FFT phase correlation with running Fourier reference + Guizar-Sicairos sub-pixel refinement, per-frame NaN-padded warp. The measured contracts stay as docstrings (Tukey-not-Hann taper, the phase magnitude floor, outlier rejection thresholds). - model.py: DriftModel without the nonrigid 'extra' carrier and the uncalled shift_at accessor. - __init__.py: no nonrigid re-exports. - drift_action.py: METHODS drops 'nonrigid'; NONRIGID_MODELS, the decimated-stack reader, the nonrigid fit step, its DEFAULTS/schema/ _coerce keys and the drift_run nonrigid branch are all gone. Non-rigid arrives in its own PR. No numba anywhere in the package (FFT/scipy/optional torch only).
…cle-movie fixture - registry: the seven drift_* staged handlers + the 'drift' wizard-schema entry (DriftWizard.parameters is the one source of truth). - toolbars.yaml: the Drift Correction toggle, gated signal_types [insitu] (the rigid solver needs a 1-D navigation axis -- the same gate Play/Fast-Forward use). - data/synthetic.py: the particle_movie generator with stamped per-frame drift ground truth (also carries the particle event table the later segmentation PR asserts against -- the fixture is shared, its identity must not fork between the two PRs). - test harness: load_test_data_particles, lazy at one frame per chunk like a real .mrc movie. The seg-only 'noise' payload knob stays behind.
… real pixels drift_wizard.spec.ts pulled from the quarry and pruned of its non-rigid stages (the 'nonrigid selectable/field controls' and 'nonrigid solve runs the fit' tests). DriftWizard.tsx pruned to the rigid tabs -- METHODS, DEFAULTS and the params() payload mirror drift_action 1:1. Wiring: FloatingToolbar mounts the caret for the Drift Correction toggle; SpyDEContext re-broadcasts the five drift_* messages as CustomEvents; protocol.ts types them.
The drift caret's Advanced disclosure changes the caret's height without re-rendering FloatingToolbar (a child's state update does not re-render its parent), so the placement went stale and the caret JUMPED on the next unrelated render -- a jump between mousedown and mouseup lands the two on different elements and the browser emits no click at all, which is how closing Advanced silently no-opped in the e2e. A ResizeObserver on the caret's box re-runs placement when its content changes; the measured width is mirrored into state so the side placements clamp into the MDI area instead of walking off the app edge. Pulled from the quarry's FloatingToolbar (minus its Segment brush strip).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Landing-order step 2: RIGID drift correction extracted from the quarry (
feat/seg-fast-engine) by the quarry rules — tests pulled first as the behavioral spec, then only the code that makes them pass, file-level pulls only, everything trimmed. Non-rigid stays behind for its own PR (nonrigid.py,test_drift_nonrigid.py, its benchmark and scripts).What landed:
spyde/drift/(frames/model/translation/warp — streaming solvers, smallDriftModeloutput, corrected movie as a LAZY per-frame view; nothing materialises the stack), the Drift Correction wizard (drift_action.py+DriftWizard.tsx+ caret wiring, nonrigid branches stripped: METHODS, models, DEFAULTS/schema keys, run branch), the 7 staged-verb registry entries + toolbars.yaml (signal_types: [insitu]), the syntheticparticle_moviefixture with stamped drift ground truth (identity shared with the future seg PR), protocol/renderer wiring, and the quarry's ResizeObserver caret-placement fix (pulled because the pruned wizard's e2e reproduced the stale-placement no-op click for real).Measured contracts kept as docstrings (the category that survives the no-narrative rule): the Tukey-not-Hann taper with the 25 px failure mode,
_PHASE_FLOOR/_REJECT_FRACTIONmeasurements, ROI-off-by-default, trace-batching flush.Evidence:
test_drift_translation.py49/49,test_drift_wizard.py57/57 (3 nonrigid tests pruned, named in the extraction log); full migrated suite 2757 passed / 0 failed (10m28s).drift_wizard.spec.ts4/4 in 33.2s (below the SLOW_SPECS bar); caret payload↔DEFAULTS↔schema verified 1:1 both directions, every setter has a caller.electron/drift_wizard_shots/01–09(caret face, Advanced with Rigid+Affine locked + backend reason, mid-fill dy/dx curve with Stop, final curve matching the fixture's 6.0 px ground truth, corrected sum visibly sharper: "1.9x sharper"). Built + headless-tested + screenshots reviewed — maintainer's eyes are the final gate on the wizard.Size, flagged honestly: 4,940 insertions over 21 files — over the 1–2k guideline. 1,573 of it is the ported tests+spec (the audited behavioral spec), 279 fixture+icon, ~3.1k production dominated by contract docstrings. If you'd rather review it in two sittings, say the word and I'll split compute (
spyde/drift/+ translation tests) from wizard (action/TSX/e2e) — the seam is clean.