Repository navigation
Adopt star-calibration v0.3.0: every pose read through its optical centre - #24
Merged
Merged
Conversation
…ical centre v0.3.0 ledger entries carry cx/cy, the optical centre the solve assumed, in pixels from the frame's middle (missing reads as 0). A solve's d_az is the azimuth of that centre, so reading it from the middle column is off by cx/k, ~1 deg at 30 px. - pose_ledger.corrected_cam attaches the solves' centre to the camera; geom reads both fisheye bearing paths (shared lens, and the whole solved camera) relative to it. - terrain_range, ridge_feet, fig_peaks, landmarks, skyline_check and the star figures project ledger and solve poses through their centre (pose_ledger.centred/solved_pose). - stars.solve solves a FIgLib sequence the way hpwren.calibrate solves a CDN block: through hpwren.camera's centre for that frame size and date, on tracks cleaned with calibrate.clean_recipe. Its batch main merges into solve_summary.json. - resolve_ledger copies CDN entries from the library's shipped ledger and re-solves only the FIgLib sequences; same sources as before. - Pin star-calibration v0.3.0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Same 86 sources. The 26 FIgLib solves are re-solved here (all still pass; 8 get an optical centre from hpwren/intrinsics.json, and their d_az moves a median 0.48 deg, max 1.18 deg on tp-w 2021-01-15); the 60 CDN solves are the library's v0.3.0 entries. Every large d_az shift tracks its centre: d_az moves by about cx/k, and corrected_cam now reads bearings from the centre, so it largely cancels at the frame's middle. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…new numbers Every stage in rerun_results.sh, from c4188b5 (out/logs/rerun_c4188b5.log), plus evolve once more after a failed Open-Meteo request dropped JunctionFire's wind. The core corpus holds: 1.87 / 1.68 / 2.02 / 2.02 km confirmed-tier upper medians, no fire moving more than 0.04 km. All-FIgLib confirmed two-site 2.02 -> 1.87 km is Rainbow alone (2.17 -> 0.99): its near-collinear bearings moved 0.06 and 0.09 deg. Terrain validate contains 62 of 74 (was 61). Ledger: residual 1.3 -> 0.9 px, consecutive nights all within 0.01 deg. 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.
Pins star-calibration v0.3.0 (tag v0.3.0, merge bf6cdb9). Ledger entries now carry
cx/cy, the optical centre's offset from the frame's middle in pixels. A missing value reads as 0.Code (2e29566)
pose_ledger.corrected_camattaches the solves' centre to the camera, andgeomreads both fisheye bearing paths relative to it. A solve's d_az is the azimuth of its centre, so reading it from the middle column would be off by about cx/k (~1° at 30 px).terrain_range,ridge_feet,fig_peaks,landmarks,skyline_checkand the star figures project poses through their centre.stars.solvesolves FIgLib nights likehpwren.calibratesolves CDN blocks: throughhpwren.camera's centre, on cleaned tracks.resolve_ledgercopies CDN entries from the library's shipped ledger.Ledger (c4188b5): rebuilt from the same 86 sources. All 26 FIgLib solves still pass, and 8 of them got a centre. Median residual is 1.3 → 0.9 px.
Results (fcb41af):
rerun_results.shfrom c4188b5, taggedresults-2026-09-30.NOTES.md (2026-09-30) has the details and the caveats: the FIgLib centres are borrowed from 2026 nights, and 18 FIgLib nights still have no centre.
🤖 Generated with Claude Code