roboflow/supervision · warning · ValueError

Unsupported metric target for IoU calculation

Error message

Unsupported metric target for IoU calculation

What it means

Inside MeanAverageRecall.compute(), after extracting per-detection content, the code dispatches IoU computation by metric target: box_iou_batch for BOXES, mask_iou_batch for MASKS, oriented_box_iou_batch for ORIENTED_BOUNDING_BOXES. This ValueError is the else-branch exhaustiveness guard: the configured _metric_target matched none of the three. Like errors 340/341 it signals a corrupted or non-enum metric_target rather than a user data problem.

Source

Thrown at src/supervision/metrics/mean_average_recall.py:455

                    prediction_confidence = np.asarray(
                        predictions.confidence, dtype=np.float32
                    )
                    if self._metric_target == MetricTarget.BOXES:
                        # BOXES target never yields CompactMask; narrow for mypy.
                        iou = box_iou_batch(
                            cast(npt.NDArray[np.number], target_contents),
                            cast(npt.NDArray[np.number], prediction_contents),
                        )
                    elif self._metric_target == MetricTarget.MASKS:
                        iou = mask_iou_batch(target_contents, prediction_contents)
                    elif self._metric_target == MetricTarget.ORIENTED_BOUNDING_BOXES:
                        # OBB target never yields CompactMask; narrow for mypy.
                        iou = oriented_box_iou_batch(
                            cast(npt.NDArray[np.number], target_contents),
                            cast(npt.NDArray[np.number], prediction_contents),
                        )
                    else:
                        raise ValueError(
                            "Unsupported metric target for IoU calculation"
                        )

                    matches, _ = _match_detection_batch_with_target_indices(
                        prediction_class_ids,
                        target_class_ids,
                        iou,
                        iou_thresholds,
                    )
                    ignored_matches = np.zeros_like(matches, dtype=bool)

                    sorted_indices = np.argsort(-prediction_confidence)
                    stats.append(
                        (
                            matches[sorted_indices],
                            ignored_matches[sorted_indices],
                            np.arange(len(prediction_confidence)),
                            prediction_class_ids[sorted_indices],

View on GitHub (pinned to 7f254d9784)

Solutions

1. Construct with a valid MetricTarget enum member: BOXES, MASKS, or ORIENTED_BOUNDING_BOXES
2. Validate config-sourced values against MetricTarget before constructing the metric
3. Do not mutate private state; create a new metric per target
4. Fork maintainers: extend the if/elif dispatch when adding a MetricTarget member

Example fix

// before
mar = MeanAverageRecall(metric_target=object())  # nonsense target
mar.compute()

// after
from supervision.metrics import MetricTarget
mar = MeanAverageRecall(metric_target=MetricTarget.BOXES)
mar.compute()

from supervision.metrics.mean_average_recall import MetricTarget
assert isinstance(metric_target, MetricTarget), 'use MetricTarget enum members only'

from supervision.metrics.mean_average_recall import MetricTarget

def is_supported_iou_target(value) -> bool:
    """True for the three targets with an IoU implementation in MAR."""
    return value in (MetricTarget.BOXES, MetricTarget.MASKS,
                     MetricTarget.ORIENTED_BOUNDING_BOXES)

try:
    mar.compute()
except ValueError as e:
    if 'Unsupported metric target' in str(e):
        raise RuntimeError('metric_target corrupted; recreate MeanAverageRecall') from e
    raise

When it happens

Trigger: An invalid metric_target value reaching the constructor (raw int, string, or foreign enum) that still passed the earlier content extraction via an unexpected path; mutating _metric_target between update() and compute(); pickling a metric across supervision versions with enum changes; custom forks adding a new MetricTarget member without extending this dispatch.

Common situations: Same class of misuse as 340/341: config-driven metric_target strings not validated; version skew between environments; fork/new-enum contributions forgetting the IoU dispatch table.

Related errors

• Invalid metric target: {self._metric_target}
• Confusion matrix must have shape (..., 3), got {confusion_ma
• Invalid metric target: {self._metric_target}
• Value must be a np.ndarray or a list
• The number of predictions ({len(predictions)}) and targets (

AI-assisted analysis of roboflow/supervision@7f254d9784 (2026-08-15). Data as JSON: /api/errors/fc1fd88552c457d0. Report an issue: GitHub.