Repository navigation
v0.4.0: sky above the skyline, centres to convergence, cross-night vouching - #3
Merged
Merged
Conversation
…s removed ablation.py refits a solve's own star-to-track matches with subsets left out (each star in turn, inner-only against outer-only by angle from the boresight, random subsets of n stars, shortened tracks), with the optical centre held or free. It also re-solves nights with tracks removed by radius or at random. hpwren/ablate.py runs both over the cache's solved blocks into ablation/ and rolls them up. solve and solve_wide take the remaining acceptance tests as parameters (max_median_px, k_window, min_tracks). The defaults are unchanged: re-solves of sample blocks match the stored results field for field. intrinsics.observations tags each point with its star. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e loop tracks: detect= passes detector options through collect, and sky_mask= replaces the top-35% band with an arbitrary sky region. detect_trial runs the big/low/sky variants over the cache; centre_iter re-solves the sky tracks through iterated per-camera centres. examples/terrain.py gains skyline and sky_mask; the San Miguel example uses the two-pass solve. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…as parameters Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ergence calibrate solve finds a block's tracks in two passes: big blobs in the top 35% for a seed pose, then the shipped blob sizes over everything above the camera's skyline (new star_calibration.terrain: Copernicus GLO-30 skyline and sky mask). calibrate intrinsics --iterate fits centres, re-solves and refits until no centre moves by more than 1 px. intrinsics.series drops a night only when its pose in the joint fit agrees with no other night, so two right nights solved through a wrong centre still refit it. detect_trial and centre_iter, the studies these came from, are retired. calibrate solve skips blocks with no frames; weather skips rows it can't place in time. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… agrees; v0.4.0 A fit turned down only for its star count (5 to 7 stars, median under 3 px, lens in its window) is solved when its pose is within 0.1 deg of the nearest night of the same camera, within 30 days, that solved on its own with 8 or more stars. calibrate.vouch decides it again from the rows after every solve; a vouched row keeps its own reason in reason_alone and names its witness in vouched_by. Version 0.4.0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…solves Fitted by calibrate intrinsics --iterate over 316 cache blocks (two-pass sky search), converged in five rounds (last move 0.0 px). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…lease READMEs and records.md: the two-pass sky search, iterated centres, vouching, the Copernicus DEM credit, and the v0.4.0 numbers against v0.3.0 on the same 308 blocks (261 solved against 242; residual 0.87 -> 0.75 px; cross-night agreement 0.022 -> 0.014 deg). 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
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.
v0.4.0: search the whole sky above the skyline, fit lens centres to convergence, and accept thin nights another night vouches for.
What changes
calibrate solve). The first pass also allows big blobs, so the brightest stars (which bloom past the old 16 px limit) find a first pose. The second keeps the usual blob sizes and searches every pixel above the camera's skyline, which a terrain model places through that pose (newstar_calibration.terrain, Copernicus DEM GLO-30). Before, only the top 35% of the frame was searched.calibrate intrinsics --iterate): fit, re-solve, refit until no centre moves by more than 1 px. A night now leaves the centre fit only if its pose in the joint fit agrees with no other night, so two right nights solved through a wrong centre still correct it.star_calibration.ablation,hpwren.ablate), and the solve's acceptance tests as parameters (defaults unchanged).intrinsics.jsonandpose_ledger.json. The detector-trial and centre-loop study modules are retired into these.Results, same 308 cache blocks
One block is lost (starr-n-mobo-c 08-21, 6 tracks). Over all 316 blocks: 267 solves on 89 cameras, now the shipped ledger.
The solves cite
0.4.0+g0157053294, the commit that ships the converged centres.Downstream
This changes which blocks solve, so plume-triangulation needs a coordinated re-run when it adopts v0.4.0. It still pins v0.2.1. No stable record field changed: result rows gain
sky,vouched_byandreason_alone, and intrinsics rows gaindropped, all documented indocs/records.md.Data: HPWREN (https://www.hpwren.ucsd.edu/). Terrain: Copernicus DEM GLO-30.
🤖 Generated with Claude Code