Function _compute_precision_and_recall

deeplabcut/core/inferenceutils.py:1091–1126

Computes the precision and recall scores at a given OKS threshold. Args: num_gt_assemblies: the number of ground truth assemblies (used to compute false negatives + true positives). oks_values: the OKS value to the matched GT assembly for each prediction oks_

(
    num_gt_assemblies: int,
    oks_values: np.ndarray,
    oks_threshold: float,
    recall_thresholds: np.ndarray,
)

def _compute_precision_and_recall(
    num_gt_assemblies: int,
    oks_values: np.ndarray,
    oks_threshold: float,
    recall_thresholds: np.ndarray,
) -> tuple[np.ndarray, np.ndarray]:
    """Computes the precision and recall scores at a given OKS threshold.

    Args:
        num_gt_assemblies: the number of ground truth assemblies (used to compute false
            negatives + true positives).
        oks_values: the OKS value to the matched GT assembly for each prediction
        oks_threshold: the OKS threshold at which recall and precision are being
            computed
        recall_thresholds: the recall thresholds to use to compute scores

    Returns:
        The precision and recall arrays at each recall threshold
    """
    tp = np.cumsum(oks_values >= oks_threshold)
    fp = np.cumsum(oks_values < oks_threshold)
    rc = tp / num_gt_assemblies
    pr = tp / (fp + tp + np.spacing(1))
    recall = rc[-1]

    # Guarantee precision decreases monotonically, see
    # https://jonathan-hui.medium.com/map-mean-average-precision-for-object-detection-45c121a31173
    for i in range(len(pr) - 1, 0, -1):
        if pr[i] > pr[i - 1]:
            pr[i - 1] = pr[i]

    inds_rc = np.searchsorted(rc, recall_thresholds, side="left")
    precision = np.zeros(inds_rc.shape)
    valid = inds_rc < len(pr)
    precision[valid] = pr[inds_rc[valid]]
    return precision, recall


def evaluate_assembly_greedy(