The fine stage searched a 5x5 neighbourhood per pyramid level and seed.
When the best coarse match is reliable (NCC score >= 0.8) and lies within
one top-level pixel of the motion prediction, a 3x3 neighbourhood is
enough; otherwise the 5x5 search stays.
Fixes#2349
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The normalized cross-correlation only had a fast path for an exactly
identity shape, which tracked footage never has after the first frame, so
every candidate went through Plane::sample. All candidates of a search are
the reference centre plus integer offsets, so each template pixel has the
same fractional offsets for all of them. Build one table of taps per
(level, seed, shape) and score each candidate with integer-indexed lerps,
with one bounding-box check up front (outside the plane the old path is
used).
lk_refine only reads the level-0 gradients around the match, so compute
them for that window (with an interior fast path in Plane::gradients_in)
and fall back to the full gradients if the iteration leaves it.
The results are unchanged: same lerp form and accumulation order.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The nightly FreeBSD run measured 119 ms/frame against a 100 ms limit: the mean included
Tracker::new and any scheduling stalls on the two-core VM. The best single step with a
250 ms bound still catches an order-of-magnitude regression.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tracker panel (Window > Tracker): Track Camera / Warp Stabilizer (disabled),
Track Motion / Stabilize Motion, Motion Source, Current Track, Track Type,
Position/Rotation/Scale, Motion Target with Edit Target..., Options...
(Motion Tracker Options dialog), Analyze (1 frame back, back, forward,
1 frame forward; stop while running) with progress, Reset and Apply (Apply
Dimensions dialog). Track point widgets in the viewer: feature and search
regions with corner handles, attach point crosshair and track path, all
draggable through track.setPoint (one undo step per drag). Automation ids for
every widget; headless UI test.
Tracker: featureless regions no longer drift (ties keep the prediction, only
confident matches predict motion). docs/agents.md: tracking example.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New L2 crate effectcraft-track: per-point feature/search region tracking with
pyramid normalized cross-correlation and Lucas-Kanade (similarity + gain/bias)
sub-pixel refinement, confidence, adapt/stop/extrapolate/continue actions,
RGB/luminance/saturation channels, blur/enhance pre-processing, and
homography/affine/similarity solves.
Project: Motion Trackers > Tracker n > Track Point n property tree
(GroupKind::Tracker with type, target and options). Engine: track.* commands
(new/motion/stabilize/property/select/setType/setTarget/options/setPoint/
analyze/stop/apply/reset/delete/status), background or blocking analysis
job with progress and cancel (one undo step), apply to Transform, Stabilize
and Corner Pin. Renderer::layer_source for the tracker's frames.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
vangel76andClaude Sonnet 5.5
Brandon ThomasandClaude Opus 5.5