Repository navigation
FIgLib nights get their own optical centres, not ones borrowed from 2026 - #25
Merged
Merged
Conversation
star_calibration's intrinsics.lookup hands a camera's nearest CDN segment to any date, so a 2020 FIgLib night took a centre fitted on 2026 frames. On lp-w-mobo-c at Willow that was (1.9, 63.1) px from one night, and it bent the solved lens (k1 -0.059) and pitch (2.4 deg). stars.intrinsics fits the FIgLib nights' own series with the library's intrinsics.build and its rules, into data/meta/figlib_intrinsics.json. A night takes its own segment's centre or the frame's middle, never a neighbour's. `python -m src.figlib.stars.intrinsics` alternates fit and re-solve twice. stars.solve.camera reads it for FIgLib sequences; CDN blocks keep the library's. solve_batch factored out of stars.solve's main. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ile; fit the FIgLib centres The first run solved every round at the frame's middle: main runs as __main__, so the copy stars.solve imports kept a stale cache. And the start matters: tp-w at Bonita matches 16 stars at the middle (under the 20-star floor, so it never gets a centre) but 22 through the library's. main now seeds each night at the library's centre, then fits and re-solves twice. 16 of 21 segments get a centre; lp-w at Willow keeps the middle (15 stars, one night). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Same 86 sources; the 60 CDN entries are unchanged. The 26 FIgLib solves move d_az a median 0.15 deg (-1.08 to +0.82) and their median residual drops 1.09 -> 0.86 px. Read at the middle column the bearings move a median 0.015 deg, at most 0.08. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every stage in rerun_results.sh from 48ae5f2 (out/logs/rerun_48ae5f2.log). Star azimuth and whole-camera rows 2.02 -> 2.03 km (Roundfire, the upper median, +0.01); terrain validate 62 -> 61 of 74 (SloaneFire lp-w back outside, as on the 2026-09-26 poses). Ledger: 0.86 px, 34 of 52 cameras more than 1 deg off, lens 0.883-0.894. star_ledger.png redrawn. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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
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.
Follow-up to #24. star-calibration's
intrinsics.lookupgives a camera's nearest CDN centre segment to any date, so a 2020 FIgLib night was taking a centre fitted on 2026 frames. On lp-w-mobo-c at Willow that borrowed centre was (1.9, 63.1) px, from a single night. It barely improved the fit (1.80 → 1.75 px), bent the lens to k1 −0.059, and moved pitch by 2.4°.Change (plume-side; the library is untouched)
stars.intrinsicsfits the FIgLib nights' own centre series with the library'sintrinsics.buildand its rules, intodata/meta/figlib_intrinsics.json.python -m src.figlib.stars.intrinsicssolves every FIgLib night through the library's centre first, then fits and re-solves twice. That starting point matters: tp-w at Bonita matches only 16 stars at the frame's middle, under the 20-star floor for fitting a centre.stars.solve.camerareads the new file for FIgLib sequences. CDN blocks keep the library's centre.Fit: 16 of 21 FIgLib camera segments get a centre. Where a camera also has a CDN centre, the two agree to within 3–8 px, so a centre usually holds across years; lp-w is the exception.
Ledger (same 86 sources; CDN entries unchanged)
Re-run (
rerun_results.shfrom 48ae5f2)🤖 Generated with Claude Code