Outputs and CSV Schema

Core Files

• Primary CSV: final trajectories for downstream analysis.
• Intermediate CSVs: forward/backward passes when enabled.
• Detection cache (.npz): optional reuse for backward/iterative runs.
• Rendered video: optional visualization output.

Common CSV Columns

The exact schema can vary by enabled features, but includes:

• TrackID: active slot identifier.
• TrajectoryID: persistent trajectory identity.
• FrameID: source frame index.
• X, Y: position in pixel coordinates.
• Theta: orientation (radians).
• State: active/occluded/lost lifecycle state.

Optional confidence-related fields (if enabled):

• DetectionConfidence
• AssignmentConfidence
• PositionUncertainty

Interpreting Output Quality

• Frequent short trajectories usually indicate weak detection or overly strict lifecycle limits.
• Large coordinate jumps suggest assignment gate or detector noise issues.
• High uncertainty with low confidence signals difficult frames worth relabeling or model retraining.

Best Practices

• Keep raw/intermediate CSVs for debugging if experimenting with thresholds.
• Use final merged CSV for analysis.
• Version output folders by date/model/parameter preset for reproducibility.