Skip to content

Feat/mirrored mode mng - #43

Merged
johnaoga merged 3 commits into
fix/config-driven-pairs-mng-datafrom
feat/mirrored-mode-mng
Sep 29, 2026
Merged

johnaoga merged 3 commits into
fix/config-driven-pairs-mng-datafrom
feat/mirrored-mode-mng

Conversation

@johnaoga

Copy link
Copy Markdown
Collaborator

No description provided.

@johnaoga
johnaoga changed the base branch from main to fix/config-driven-pairs-mng-data September 17, 2026 09:49
@johnaoga
johnaoga added this pull request to stack #44 September 17, 2026 09:50
johnaoga and others added 3 commits September 29, 2026 12:49
… seed gates, frame cache

Follow-ups from the 2026-07-22 real-data smoke test (MNG rig, Zhaochong index
finger), rebased on the config fixes:

- run(config, trials): honours --trial/--trials (default: ref_trial_id only).
- processPair() is now wrapped in try/catch: an engine exception in one pair
  ('could not seed any current image') no longer aborts the other 6 pairs.
- runViewDic() mirrors StepDWorkflow::runNcorrAnalysis: parallel
  (matlab_DIC_analysis_parallel, step_d_total_threads) when
  parallel_processing, NO_UPDATE preset via no_update, fixed_step_ref,
  seeds_are_optimized, cutoff_max_diffnorm and the seed-quality gate. The
  view1->view2 matching call uses matching_cutoff_max_corrcoef (1.0): on the
  smoke-test data the matching seed converges at corrcoef 0.51 and the 0.5
  tracking gate rejected it — that, not a missing algorithm, was the July
  failure (verified with proxyncorr on the extracted view halves; a flipped
  view2 diverges, so the views are not mirror-reversed).
- checkpoints: existing ncorr{v}.bin / ncorr{v1}{v2}.bin are reloaded.
- saturation honours im_saturation_mode / limit_grayscale (MNG default 120).
- per-(trial, view) frame cache under tmp_frames_mirrored/T<t>/view<n> with a
  window signature marker, so views shared by adjacent pairs decode once.
- deploy: cppxdic_mirrored.def + submit_job_mirrored.sh (July assets) added.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Same fallback as the camerapairs matchingInitialFrame (205faeb).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
3 of the 4 masked pairs (1, 5, 7) now reconstruct end to end at 700 frames,
against 0 of 7 in the July smoke test. 2h46 / 35.7 GB per run on 32 threads;
outputs byte-identical between keira and a gpu-partition node. Remaining
blockers are data-side (no REF mask/seed for pairs 2/4/6, pair 3 matching) plus
the unported MNG Step E.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@johnaoga
johnaoga force-pushed the feat/mirrored-mode-mng branch from 2eb14a0 to 7b282cf Compare September 29, 2026 10:49
@johnaoga
johnaoga merged commit ec718da into main Sep 29, 2026
4 checks passed
An error occurred while trying to automatically change base from fix/config-driven-pairs-mng-data to main September 29, 2026 10:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant