Repository navigation
Sky model checked against the terrain; terrain range and recent poses re-run - #18
Merged
Merged
Conversation
…r on any ledger - stars/skyline_check.py: re-solve each camera's night under five sky models and score the DEM skyline against the image's own ridge edge (10-78 km ridges), plus the two figures. - geolocate.refine / FIGLIB_REFINE_KM: re-find the likelihood peak at 10 m within two grid cells, so it no longer moves with the grid. Opt-in, and terrain_range uses it the same way, so published numbers stay on the grid. - resolve_ledger takes a ledger path; entries the tracked ledger already has under the current sky model are copied, not solved again. - ridge_feet figure: the hp-e caption reads the measured shift. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- docs/star-calibration.md "The terrain agrees": on 39 cameras the DEM skyline's median offset from the ridge goes -4.2 -> +0.3 px, and the mean |offset| 6.4 -> 2.8 px, with precession doing most of it. This explains the 2026-09-13 skyline bias (ridge_feet edges -5.5 -> -1.0 px). - README terrain section re-run: validation 60/73 as before. Rainbow's bearings alone go 5.82 -> 1.42 km, so terrain now only shrinks its region (25.6 -> 4.8 km2). Palisades no longer moves (1.82 km), and its figure is dropped. The 30 px band now hurts the confirmed median (1.82 -> 2.00). - NOTES: the skyline check, the grid-resolution study (the ledger variant's confirmed median is ~2.0 km; the counts within 2 km are fragile), the terrain re-run, and the recent-corpus re-solve (Bernardo 0.63 and Junction 3.54 km unchanged; bm-e 09-19 and sojr-s fall under the gates). 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.
Stacked on #17. Retarget to
mainafter #17 merges.Skyline check (
stars/skyline_check.py, new section in docs/star-calibration.md). Each of 41 cameras' nights is re-solved under five sky models. The DEM skyline under each pose is then compared with the ridge edge in a daytime frame from the same day.35 of 39 cameras move closer to the ridge. This also explains the 2026-09-13 skyline bias NOTES called "mostly unexplained":
ridge_feet edgesgoes from −5.5 to −1.0 px.Grid resolution. The ledger variant's confirmed median really is about 2.0 km, not 1.7. With the peak refined off the grid it is 1.87 / 1.68 / 2.02 km for baseline / fisheye / ledger. The counts within 2 km are the fragile number.
geolocate.refineis added as opt-in (FIGLIB_REFINE_KM), because turning it on moves every km figure in the README by up to ~0.3 km (Kitchen 0.08 → 0.14). That should be one coordinated re-run, not part of this PR.Terrain range, re-run on the precessed ledger:
Recent corpus:
resolve_ledgernow takes a ledger path. 14 of the 16 extra solves re-solve. bm-e 09-19 and sojr-s 09-20 fall just under the solver gates. The Bernardo (0.63 km) and Junction (3.54 km) scores are unchanged.Found on the way, not fixed:
terrain_rangesetsFIGLIB_LENS=fisheyeat import. As a resulttests/test_geolocation.pyfails its two-site test when run alone, on main too.🤖 Generated with Claude Code