Skip to content

boxmot.engine.calibration.cmc

Image-only camera analysis shared by joint camera/Kalman selection candidates.

PreparedCMCCandidates dataclass 

PreparedCMCCandidates(
    warps: Mapping[str, ndarray],
    policy: SequenceCMCPolicy,
    report_path: Path,
    warps_path: Path,
    report: Mapping[str, Any],
    mask_min_conf: float,
)

Frozen camera behavior reused for every filter, profile, and fold.

warps[sequence][frame] maps the preceding delivered frame to the current frame, with identity at frame zero and on disabled sequences. The NPZ artifact retains raw estimates even for disabled sequences.

mask_min_conf instance-attribute 

mask_min_conf: float

policy instance-attribute 

policy: SequenceCMCPolicy

report instance-attribute 

report: Mapping[str, Any]

report_path instance-attribute 

report_path: Path

warps instance-attribute 

warps: Mapping[str, ndarray]

warps_path instance-attribute 

warps_path: Path

cmc_analysis_settings 

cmc_analysis_settings(
    options: Mapping[str, Any] | None = None,
) -> dict[str, Any]

Describe the resolved SOF estimator and sequence gate in saved JSON.

prepare_cmc_candidates 

prepare_cmc_candidates(
    args: Namespace,
    baseline: Mapping[str, Any],
    output_dir: str | Path,
    progress: Callable[[str], None] | None = None,
) -> PreparedCMCCandidates

Prepare image-only off and automatic SOF behavior for joint selection.

args must already contain the evaluator's resolved build_path, split, seq_info, and optional sequence_names. No ground-truth coordinates, tracker predictions, or tracking metrics enter the decision. The caller freezes the returned masking threshold in subsequent tuning.