Add decoder motion vector export and experimental VideoRegistration module - #463
Merged
Conversation
Motion vector extraction (src/avio.jl): - `openvideo(file; export_mvs=true)` sets AV_CODEC_FLAG2_EXPORT_MVS so the decoder exports block motion vectors as frame side data (software decoding only; errors if combined with `hwaccel`) - `motion_vectors(reader)` returns the `MotionVector`s of the most recently read frame, mirroring `frame_metadata`; side data is captured at decode time into a FIFO kept in lockstep with the frame queue across read/skip/seek - sub-pixel helpers `displacement`, `src_position`, `dst_position`, and `correspondences(mvs)` converting vectors to filtered (dst, src) point-pair matrices VideoIO.VideoRegistration (src/registration.jl): - self-contained, purely point-based robust 2D registration module (stdlib deps only) designed to be movable to a standalone package unchanged; all codec-specific knowledge stays in VideoIO, with the motion-vector convenience method as glue in src/registration_glue.jl - closed-form translation / similarity (Umeyama) / affine fits wrapped in RANSAC with adaptive early termination and inlier refinement - `estimate_global_motion` -> `GlobalMotion` with quality metrics (inlier fraction, median residual, spatial coverage), `local_residuals` for residual local motion, and `classify_confidence` implementing a use-directly / initialize-refinement / full-registration-fallback decision rule Also: util/motion_vector_demo.jl renders a side-by-side annotated video (inlier/outlier dots, global-translation arrow); docs pages, tests (round-trip against synthetic H.264 with known ground-truth motion), and CHANGELOG entry; version bumped to 1.9.0. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #463 +/- ##
=========================================
Coverage ? 81.63%
=========================================
Files ? 12
Lines ? 1721
Branches ? 0
=========================================
Hits ? 1405
Misses ? 316
Partials ? 0 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
…ion experimental Demo (util/motion_vector_demo.jl): - third pane showing the frame corrected for global movement, warped by the chained per-frame estimates with bilinear sampling - --smooth=N: correct only the deviation from an exponentially smoothed trajectory instead of a fixed anchor (follows slow/intentional motion, immune to long-sequence drift) - --refine: hybrid image-domain SAD refinement seeded by the codec estimate, with parabolic sub-pixel interpolation. Recovers global motion that encoders code as residuals instead of motion vectors on low-contrast or grainy content (where near-all blocks are zero-motion skips and form a false consensus), and bridges I-frames, which carry no vectors - --coast / --model=<m> chain-policy flags - grain-robust temporal-stability metric (downsampled central crop) and chain-drift indicator printed per run On the Annie Oakley test video the hybrid refine+smooth settings reach the measured geometric ceiling (pairwise-optimal alignment); the residual frame difference is dominated by luminance flicker and film grain, which geometric registration cannot remove. VideoRegistration: - add compose_transform for chaining per-frame estimates into a cumulative anchor-relative transform (exported, documented, tested) - mark the module experimental in the module docstring, docs page, and CHANGELOG: API may change between minor releases and the module may move to a separate package (it is self-contained to permit that); the extraction API in VideoIO proper is unaffected Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
IanButterworth
force-pushed
the
motion-vector-registration
branch
from
August 7, 2026 01:42
b5e1f25 to
7509b1c
Compare
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.
Adds a way to extract any already available motion information from the video file during read.
Also adds an experimental VideoRegistration module, which may be moved to a separate package.
Demoing it with
util/motion_vector_demo.jlThe one on the right is an opinionated rough registration based on the info. There are many ways to use the data
mv_demo_synthetic.mp4
mv_demo_annie_oakley.mp4