Batch of small fixes - #1007
Merged
Merged
Conversation
A falsy check dropped camera 0 into anchor auto-selection. Guard on 'is not None' instead. Regression test hand-builds a three-camera network where auto-selection prefers a different anchor, so the test fails without the fix.
All epipolar scenes used contiguous ids 0..n-1, so cam_id/index
conflation could pass unnoticed. New scene drops cameras 0, 3, 4 from
a six-ring, leaving ids {1, 2, 5}, and drives the 2D-only path through
the production seam. Measured worst error 0.062 deg / 1.14 mm against
0.5 deg / 10 mm bounds.
The floor was the single minimum world Z, so one spurious low triangulation buried it and floated the whole scene. Use the 1st percentile with method='lower': an order statistic that degrades to the exact minimum on small point sets, leaving existing behavior and tests unchanged, while skipping stray low points on dense clouds.
The consensus vertical averaged per-camera ups and discarded how much they disagreed, though cross-camera agreement is the accuracy signal for the vertical. Log each camera's angle from the consensus and the max pairwise disagreement in degrees. Log-only; the rotation is unchanged. Also name the module logger by __name__ instead of __file__, matching the rest of the codebase, so subscribers can find it under caliscope.core.capture_volume.
The signal had no connect sites anywhere in src or tests; every emit went unheard. Remove the signal, push_camera_data, its two call sites, and the stale docstring sentence. CameraData.get_display_data lost its only caller, so it goes too, along with the orphaned OrderedDict imports in both files.
mprib
changed the title
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.
Five small fixes from the July 15 review round. One commit each.
Camera 0 anchor treated as unset
apply_tochecked the anchor camera for truthiness (if anchor_cam:), so camera 0 read as unset and fell through to auto-selection (src/caliscope/core/bootstrap_pose/paired_pose_network.py:195). Fixed tois not None. The regression test builds a three-camera network where auto-selection picks a different anchor than camera 0. The test fails without the fix.Non-contiguous camera ID canary
Every epipolar test scene used contiguous IDs 0 through n-1. That let code silently confuse camera IDs with array indices. The new scene drops cameras 0, 3, and 4 from a six-camera ring, leaving IDs {1, 2, 5}, and runs the 2D-only path through the production seam. Measured worst error: 0.062 deg / 1.14 mm against bounds of 0.5 deg / 10 mm.
Robust floor in grounded()
The floor was the single minimum world-Z value. One spurious low triangulation could bury it. Now it uses the 1st percentile as an order statistic (
np.percentile(z, 1.0, method="lower")). On small point sets this degrades to the exact minimum. The existing two-point grounding test passes unchanged. New tests: a 200-point band with one low outlier rests at ~0 after grounding. A clean dense volume still grounds to its true minimum.Cross-camera vertical disagreement in oriented()
The consensus vertical averaged per-camera up vectors but discarded how much they disagreed. Cross-camera agreement is the accuracy signal. Each camera's angle from the consensus and the max pairwise disagreement are now logged in degrees. This is log-only. The rotation is unchanged. The module logger is also named by
__name__instead of__file__, matching the other 75 modules so subscribers can find it.Dead new_camera_data chain deleted
The signal had zero connect sites in src or tests. Removed the signal,
push_camera_data, its two call sites, the stale docstring sentence,CameraData.get_display_data(whose only caller was the deleted helper), and the orphanedOrderedDictimports. 95 lines removed. No behavior change.