Add a reusable video alignment module for offline multiview workflows. The helper scans per-frame timestamps, builds nearest-timestamp bundle matches under a configurable skew threshold, and rewrites synchronized per-camera videos for downstream detection and tracking runs.
The detection package now exports the alignment primitives, and a test-support CLI is included so dataset-specific experiments can generate aligned clips without expanding the public application surface.
Regression tests cover both bundle matching and frame selection during rewritten video generation.
Add an explicit source delivery policy to the detection pipeline so offline and realtime sources can be handled differently without splitting the runner.
Blocking sources now backpressure ingestion until their pending frame is drained, which preserves every offline video frame even when inference is slower than decode. Latest-only sources keep the previous overwrite behavior for realtime feeds such as cvmmap.
The tests now cover both policies: offline sources retain ordered frame delivery under slow inference, while latest-only sources still drop intermediate frames as intended.
Add an idempotent patch-mmdet-version-gate entrypoint that rewrites the installed mmdet mmcv compatibility assert into a warning so the local rebuilt mmcv wheel can be reused after uv sync.
Cover the source rewrite with focused tests and document the post-sync command in the README so the local environment patch is reproducible instead of being a one-off manual edit inside .venv.
Pin the detection stack to torch 2.11.0 and torchvision 0.26.0 and route both packages through the explicit PyTorch cu130 index for Linux and Windows installs.
Move the remaining uv project configuration into pyproject.toml, set PyPI as the default index, and remove the redundant repo-local uv.toml so uv no longer emits split-configuration warnings.
Refresh the lockfile against the new index configuration so the project resolves from pypi.org for general packages and download.pytorch.org/cu130 for the CUDA-specific torch artifacts.
Refactor the package into common, schema, detection, and tracking namespaces and move dataset-specific ActualTest utilities into tests/support.
Add a pluggable detection stack with typed protocols, pydantic-settings config, loguru-based runner logging, cvmmap and headless video sources, NATS and parquet sinks, and a structured coco-wholebody133 payload path.
Teach tracking replay loading to consume parquet detection directories directly, preserve empty frames, and keep the video-to-parquet-to-tracking workflow usable for offline E2E runs.
Vendor the local mmcv and xtcocotools wheels under Git LFS, update uv sources/lock state, and refresh the mmcv build so mmcv.ops loads successfully with the current torch+cu130 environment.
Set up pose_tracking_exp as a uv-managed Python package for offline multiview body tracking experiments.
This initial commit includes:
- the typed package scaffold, CLI entrypoints, and repo-local uv configuration
- scene and replay loaders for generic JSON replays and ActualTest parquet inputs
- ParaJumping payload conversion and RTMPose-to-body20 normalization
- a custom articulated tracker with tentative, active, and lost lifecycle handling
- RPT-backed proposal generation, camera convention handling, and multiview reprojection updates
- regression tests for normalization, camera conventions, ActualTest ingestion, seeding, and tracker smoke flows
- project documentation covering extrinsic formats and the ActualTest calibration caveat
crosstyan