roboflow/supervision · error · ValueError

confidence must be a 1D np.ndarray with shape {expected_shap

Error message

confidence must be a 1D np.ndarray with shape {expected_shape}, but got shape {actual_shape}

What it means

Raised by supervision.validators._validate_confidence when confidence is given to Detections but is neither None nor a 1D np.ndarray of shape (n,) matching the number of boxes. Confidence holds per-detection scores in [0, 1].

Source

Thrown at src/supervision/validators/__init__.py:114

@deprecated(  # type: ignore[untyped-decorator]
    target=_validate_class_id,
    deprecated_in="0.29.0",
    remove_in="0.32.0",
)
def validate_class_id(class_id: Any, n: int) -> None:
    void(class_id, n)


def _validate_confidence(confidence: Any, n: int) -> None:
    """Validate detection-level confidence: 1D ``np.ndarray`` with shape ``(n,)``."""
    expected_shape = f"({n},)"
    actual_shape = str(getattr(confidence, "shape", None))
    is_valid = confidence is None or (
        isinstance(confidence, np.ndarray) and confidence.shape == (n,)
    )
    if not is_valid:
        raise ValueError(
            f"confidence must be a 1D np.ndarray with shape {expected_shape}, but got "
            f"shape {actual_shape}"
        )


@deprecated(  # type: ignore[untyped-decorator]
    target=_validate_confidence,
    deprecated_in="0.29.0",
    remove_in="0.32.0",
)
def validate_confidence(confidence: Any, n: int) -> None:
    void(confidence, n)


def _validate_keypoint_confidence(confidence: Any, n: int, m: int) -> None:
    """Validate per-keypoint confidence: 2D ``np.ndarray`` with shape ``(n, m)``."""
    actual_shape = str(getattr(confidence, "shape", None))

View on GitHub (pinned to 7f254d9784)

Solutions

1. Convert to a flat NumPy array: confidence=np.asarray(scores).ravel().
2. Apply the same boolean mask used on xyxy to confidence so lengths stay equal.
3. Convert Torch tensors first: scores.detach().cpu().numpy().
4. Omit confidence (None) if your detector does not produce scores.

Example fix

# before
dets = Detections(xyxy=boxes, confidence=[[0.9], [0.8]])  # (2,1) -> ValueError

# after
dets = Detections(xyxy=boxes, confidence=np.array([0.9, 0.8]))

n = len(xyxy)
confidence = None if confidence is None else np.asarray(confidence, dtype=np.float32).ravel()
assert confidence is None or confidence.shape == (n,)
dets = Detections(xyxy=xyxy, confidence=confidence)

def is_valid_confidence(confidence: Any, n: int) -> bool:
    return confidence is None or (
        isinstance(confidence, np.ndarray) and confidence.shape == (n,)
    )

When it happens

Trigger: Passing confidence as a Python list, a 2D array like (n, 1), or an array of length different from len(xyxy) when constructing Detections.

Common situations: Taking raw model output scores that come as (n, 1) and passing them unmodified; mixing Torch tensors (convert with .cpu().numpy()) or lists instead of NumPy arrays; filtering boxes without filtering confidence the same way.

Related errors

• xyxy must be a 2D np.ndarray with shape {expected_shape}, bu
• class_id must be a 1D np.ndarray with shape {expected_shape}
• tracker_id must be a 1D np.ndarray with shape {expected_shap
• Detections must have class_id attribute.
• Both Detections should have exactly 1 detected object.

AI-assisted analysis of roboflow/supervision@7f254d9784 (2026-08-15). Data as JSON: /api/errors/c48fb13aac3e1846. Report an issue: GitHub.