keras-team/keras · error · ValueError

Found bounding_boxes['boxes'].shape={boxes_shape} and expect

Error message

Found bounding_boxes['boxes'].shape={boxes_shape} and expected bounding_boxes['labels'] to have rank 2 or 3, but received: bounding_boxes['labels'].shape={labels_shape} 

What it means

For dense tensors, when boxes has rank 3 (batched, shape (batch, num_boxes, 4)), labels must have rank 2 (batch, num_boxes) or rank 3 (batch, num_boxes, num_classes). Any other rank breaks the per-batch, per-box label alignment.

Source

Thrown at keras/src/layers/preprocessing/image_preprocessing/bounding_boxes/validation.py:168

                " `bounding_boxes['labels']` must also be a "
                "Ragged tensor. "
                f"Received: bounding_boxes['labels']={labels}"
            )
    else:
        boxes_shape = current_backend.shape(boxes)
        labels_shape = current_backend.shape(labels)
        if len(boxes_shape) == 2:  # (boxes, 4)
            if len(labels_shape) not in {1, 2}:
                raise ValueError(
                    "Found "
                    f"bounding_boxes['boxes'].shape={boxes_shape} "
                    "and expected bounding_boxes['labels'] to have "
                    "rank 1 or 2, but received: "
                    f"bounding_boxes['labels'].shape={labels_shape} "
                )
        elif len(boxes_shape) == 3:
            if len(labels_shape) not in {2, 3}:
                raise ValueError(
                    "Found "
                    f"bounding_boxes['boxes'].shape={boxes_shape} "
                    "and expected bounding_boxes['labels'] to have "
                    "rank 2 or 3, but received: "
                    f"bounding_boxes['labels'].shape={labels_shape} "
                )
        else:
            raise ValueError(
                "Expected `bounding_boxes['boxes']` "
                "to have rank 2 or 3, with shape "
                "(num_boxes, 4) or (batch_size, num_boxes, 4). "
                "Received: "
                f"bounding_boxes['boxes'].shape={boxes_shape}"
            )

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Reshape labels to (batch, num_boxes): labels.reshape(boxes.shape[0], boxes.shape[1])
2. For one-hot labels use (batch, num_boxes, num_classes)
3. Verify batch dim of labels equals boxes.shape[0]

Example fix

# before
bbs = {'boxes': boxes, 'labels': labels.reshape(-1)}
# after
bbs = {'boxes': boxes, 'labels': labels.reshape(boxes.shape[0], boxes.shape[1])}

if len(bbs['boxes'].shape) == 3:
    assert len(bbs['labels'].shape) in (2, 3), bbs['labels'].shape

When it happens

Trigger: densify_bounding_boxes with boxes shape (8, 10, 4) but labels of rank 1 or rank 4, e.g. flat labels of shape (80,).

Common situations: Forgetting to batch labels when moving from single-image to batched inference; flattening labels with reshape(-1); keeping a trailing singleton dim making rank 4.

Related errors

• Found bounding_boxes['boxes'].shape={boxes_shape} and expect
• Expected `bounding_boxes['boxes']` to have rank 2 or 3, with
• `height` and `width` must be set if `format='xyxy'`.
• `variance` must be length 4, got {variance}
• `encoding_format` should be one of 'center_xywh' or 'center_

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/4e979d18e464b1f8. Report an issue: GitHub.