huggingface/transformers · error · ValueError

Unsupported input type {type(bboxes_center)}

Error message

Unsupported input type {type(bboxes_center)}

What it means

center_to_corners_format dispatches on input type: torch tensors go to the torch kernel, numpy arrays to the numpy kernel; anything else (lists, tuples) raises ValueError. Used in detection forward passes, so it avoids silent device round-trips by refusing unknown types.

Source

Thrown at src/transformers/image_transforms.py:565

# 2 functions below inspired by https://github.com/facebookresearch/detr/blob/master/util/box_ops.py
def center_to_corners_format(bboxes_center: TensorType) -> TensorType:
    """
    Converts bounding boxes from center format to corners format.

    center format: contains the coordinate for the center of the box and its width, height dimensions
        (center_x, center_y, width, height)
    corners format: contains the coordinates for the top-left and bottom-right corners of the box
        (top_left_x, top_left_y, bottom_right_x, bottom_right_y)
    """
    # Function is used during model forward pass, so we use torch if relevant, without converting to numpy
    if is_torch_tensor(bboxes_center):
        return _center_to_corners_format_torch(bboxes_center)
    elif isinstance(bboxes_center, np.ndarray):
        return _center_to_corners_format_numpy(bboxes_center)

    raise ValueError(f"Unsupported input type {type(bboxes_center)}")


def _corners_to_center_format_torch(bboxes_corners: "torch.Tensor") -> "torch.Tensor":
    top_left_x, top_left_y, bottom_right_x, bottom_right_y = bboxes_corners.unbind(-1)
    b = [
        (top_left_x + bottom_right_x) / 2,  # center x
        (top_left_y + bottom_right_y) / 2,  # center y
        (bottom_right_x - top_left_x),  # width
        (bottom_right_y - top_left_y),  # height
    ]
    return torch.stack(b, dim=-1)


def _corners_to_center_format_numpy(bboxes_corners: np.ndarray) -> np.ndarray:
    top_left_x, top_left_y, bottom_right_x, bottom_right_y = bboxes_corners.T
    bboxes_center = np.stack(
        [
            (top_left_x + bottom_right_x) / 2,  # center x

View on GitHub (pinned to a597f97485)

Solutions

1. Convert lists to np.ndarray: np.asarray(boxes).
2. Keep boxes as whatever your framework kernel produced (torch tensors stay torch).
3. Ensure shape is (..., 4) center format before converting.

Example fix

# before
corners = center_to_corners_format([[50.0, 50.0, 100.0, 100.0]])  # raises

# after
import numpy as np
corners = center_to_corners_format(np.array([[50.0, 50.0, 100.0, 100.0]]))

import numpy as np
if not (is_torch_tensor(bboxes_center) or isinstance(bboxes_center, np.ndarray)):
    bboxes_center = np.asarray(bboxes_center, dtype=float)
assert bboxes_center.shape[-1] == 4

def is_supported_boxes(x) -> bool:
    import numpy as np
    from transformers.utils import is_torch_tensor
    return is_torch_tensor(x) or isinstance(x, np.ndarray)

When it happens

Trigger: center_to_corners_format([[50, 50, 100, 100]]) with a plain Python list, or passing a tf.Tensor / jax array.

Common situations: Feeding model outputs or config values that are plain lists into detection postprocessing (e.g. post_process_object_detection outputs are fine, but hand-built boxes are not), or mixing frameworks.

Related errors

• Unsupported input type {type(bboxes_corners)}
• `axis_value` for `HQQ` backend has to be one of [`0`, `1`] b
• You can construct a Cache either from a list `layers` of all
• You should provide exactly one of `layers` or `layer_class_t
• Cannot call `update_conv_state` on a non-LinearAttention lay

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/e5e251089be0990b. Report an issue: GitHub.