Fixes for strain mapping - #271
Open
cophus wants to merge 141 commits into
Open
Conversation
…_hotfix Tomography Dataset Hotfix
…re/update-lockfile-20260629 chore: Update dependencies
…re/update-lockfile-20260706 chore: Update dependencies
create_batch_rays and integrate_rays were decorated with @torch.compile(mode="reduce-overhead"). Calling them triggers TorchInductor, which needs a C compiler to build its kernel. On Windows runners without MSVC (cl.exe) on PATH this raises InductorError, failing tests/tomography/test_dataset_models.py::TestINRRayMath on windows-latest. Both are trivial tensor ops (a small fill and a single reduction) and reduce-overhead is a no-op on CPU, so eager execution is equivalent and portable across platforms.
…-torch-compile-rays Remove torch.compile from INR ray helpers for Windows compatibility
…ease/0.1.9 Bump version to 0.1.9
…re/update-lockfile-20260713 chore: Update dependencies
Sync main to dev
…re/update-lockfile-20260720 chore: Update dependencies
…re/update-lockfile-20260727 chore: Update dependencies
get_individual_uv_vectors fell through to pos_state.keys() after writing NaNs for an unfitted position, raising AttributeError whenever fit_individual_diffraction_pattern was run with rows=/cols= subsets; skip to the next position instead. reset(reset_to="individual") now raises a clear RuntimeError for an unfitted position rather than failing inside state loading. StrainMapAutocorrelation.diffraction_mask raised a zero-size-array ValueError when edge_blend (default 64) exceeded the kept region's inradius (e.g. any 128x128 detector); clamp edge_blend to the largest feasible feather with a warning, and raise a clear error when the threshold masks every pixel. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
probe_centroid built its index grids on the CPU, crashing make_template_from_data on a CUDA/MPS device. The cepstral batched fitter forced float64, which MPS does not support at all and consumer CUDA cards run ~30x slower than float32; float32 is ample for the ~0.01 px peak-fit noise floor (verified identical strain output on the LFP tutorial dataset). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The correlation map is relu-clamped before the local-maxima search. With a zero-sum template, a bright unscattered beam imprints a broad negative moat on the correlation map; weak disk peaks ride below zero there and the clamp erased them before any threshold applied -- at quasi-vacuum positions of the LFP tutorial dataset 96% of the map clamped to zero and only 4 of 39 maxima survived, and even on-particle patterns lost their weakest disks. Add background_sigma: a Fourier-domain Gaussian high-pass applied to the correlation product (batched, both fft and rfft paths, also used by the DFT subpixel refinement) that removes the smooth background so peak intensities become prominence above the local baseline. BraggVectors detect_disks / show_detection / correlation_map default to background_sigma="auto" (twice the central-beam radius); pass None for the previous behavior. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Design decision: keep the correlation map strictly positive so peak intensities roughly measure the probe-weighted counts under each disk and min_abs_intensity is always a nonnegative counts-scale threshold. The previous zero-sum default subtracted the pattern-mean from the correlation, pushing weak disk peaks below zero where the relu clamp erased them (the missing-peaks bug on the LFP tutorial data). make_template_* now default subtract_mean=False; background_sigma (the Gaussian high-pass) defaults to None everywhere, remaining available for patterns with a strong smooth background. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
One-call utility that shifts the detected peak coordinates so every scan position shares a common origin: subtracts the supplied per-position diffraction origins (e.g. the plane-fitted CoM of the central beam) and adds back a reference origin (default: the scan mean), then recomputes the BVM. The raw data is never resampled -- the measurements are calibrated, not the images. Returns a corrected copy by default so repeated calls (re-running a notebook cell) never double-apply the shift; the shift itself is applied vectorized on the flat peak table via Vector.flatten()/set_flattened(). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ModelDiffraction.preprocess(origins=...) builds a descan-free mean CBED: each pattern is shifted by (scan-mean origin - its measured origin), e.g. from the plane-fitted CoM of the central beam, before averaging. Only the image_ref fit target is built from shifted copies -- the raw data is never resampled. The existing align=True cross-correlation self-alignment stays available but is unreliable when diffraction contrast varies across the scan (on the LFP tutorial data it produced +/-68 px shifts against a true +/-5 px descan, chasing pattern content instead of the beam); origins takes precedence when both are given. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… model fitting, and optimizer batched implementation
…odels' into strain-fitting-multi-optimizer fixing some conflicts
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
7 participants
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.
Some bug fixes for disk detection strain mapping.