Data Flow

MAT Pipeline

1. Video frames are read (optionally prefetched).
2. Detector generates per-frame measurements.
3. Kalman prediction and assignment update track state.
4. Worker emits frame/status/metrics to GUI.
5. Trajectories are written to CSV.
6. Optional backward pass reuses detection cache.
7. Post-processing resolves and interpolates trajectories.

Key Data Artifacts

• Measurements: detector outputs including center, orientation, size cues.
• Track State: predicted/corrected state vectors and covariance.
• Cache: .npz detection cache for repeatability and performance.
• CSV: final analysis artifact for downstream pipelines.

PoseKit Pipeline

1. Image set + project metadata loaded.
2. Annotation state edited in UI.
3. Labels persisted to YOLO pose format.
4. Optional model-assisted inference and split-generation steps create derived artifacts.