facebookresearch/detectron2 · error · NotImplementedError

Conversion from BoxMode {} to {} is not supported yet

Error message

Conversion from BoxMode {} to {} is not supported yet

What it means

BoxMode.convert implements only XYXY_ABS<->XYWH_ABS and to/from XYWHA_REL (partial) conversions. Any other pair, notably anything involving XYWHA_ABS or XYWHA_REL with absolute inputs in unsupported directions, raises NotImplementedError.

Source

Thrown at detectron2/structures/boxes.py:116

            arr[:, 3] = arr[:, 1] + new_h

            arr = arr[:, :4].to(dtype=original_dtype)
        elif from_mode == BoxMode.XYWH_ABS and to_mode == BoxMode.XYWHA_ABS:
            original_dtype = arr.dtype
            arr = arr.double()
            arr[:, 0] += arr[:, 2] / 2.0
            arr[:, 1] += arr[:, 3] / 2.0
            angles = torch.zeros((arr.shape[0], 1), dtype=arr.dtype)
            arr = torch.cat((arr, angles), axis=1).to(dtype=original_dtype)
        else:
            if to_mode == BoxMode.XYXY_ABS and from_mode == BoxMode.XYWH_ABS:
                arr[:, 2] += arr[:, 0]
                arr[:, 3] += arr[:, 1]
            elif from_mode == BoxMode.XYXY_ABS and to_mode == BoxMode.XYWH_ABS:
                arr[:, 2] -= arr[:, 0]
                arr[:, 3] -= arr[:, 1]
            else:
                raise NotImplementedError(
                    "Conversion from BoxMode {} to {} is not supported yet".format(
                        from_mode, to_mode
                    )
                )

        if single_box:
            return original_type(arr.flatten().tolist())
        if is_numpy:
            return arr.numpy()
        else:
            return arr


class Boxes:
    """
    This structure stores a list of boxes as a Nx4 torch.Tensor.
    It supports some common methods about boxes
    (`area`, `clip`, `nonempty`, etc),

View on GitHub (pinned to a2f4a8771a)

Solutions

1. Convert manually (e.g. with shapely or box_ops) for the unsupported pair
2. Convert to an intermediate supported mode (XYXY_ABS or XYWH_ABS) in two steps if direction is the issue
3. Use detectron2.structures.rotated_boxes.rotated_box_to_aligned_box for XYWHA_ABS -> XYXY_ABS

Example fix

# before
boxes_xyxy = BoxMode.convert(boxes, BoxMode.XYWHA_ABS, BoxMode.XYXY_ABS)
# after
from detectron2.structures.rotated_boxes import rotated_box_to_aligned_box
boxes_xyxy = rotated_box_to_aligned_box(boxes)

from detectron2.structures import BoxMode
supported = {(BoxMode.XYXY_ABS, BoxMode.XYWH_ABS), (BoxMode.XYWH_ABS, BoxMode.XYXY_ABS), (BoxMode.XYXY_ABS, BoxMode.XYWHA_REL), (BoxMode.XYWHA_REL, BoxMode.XYXY_ABS), (BoxMode.XYWH_ABS, BoxMode.XYWHA_REL)}
assert (from_mode, to_mode) in supported, 'unsupported BoxMode conversion'

def can_convert(f, t) -> bool:
    return (f, t) in {(BoxMode.XYXY_ABS, BoxMode.XYWH_ABS), (BoxMode.XYWH_ABS, BoxMode.XYXY_ABS), (BoxMode.XYXY_ABS, BoxMode.XYWHA_REL), (BoxMode.XYWHA_REL, BoxMode.XYXY_ABS), (BoxMode.XYWH_ABS, BoxMode.XYWHA_REL)}

try:
    out = BoxMode.convert(b, f, t)
except NotImplementedError:
    out = manual_convert(b, f, t)  # e.g. rotated_box_to_aligned_box

When it happens

Trigger: Calling BoxMode.convert(boxes, from_mode=BoxMode.XYWHA_ABS, to_mode=BoxMode.XYXY_ABS); the rotated-box branch only handles a subset of direction/mode pairs.

Common situations: Rotated detection (RRPN/Rotated BBox datasets like DOTA) converting rotated annotations to axis-aligned boxes; mixing REL and ABS conventions across datasets.

Related errors

• Cannot match one checkpoint key to multiple keys in the mode
• Class with @configurable must have a 'from_config' classmeth
• {name} must take 'cfg' as the first argument!
• target of LazyCall must be a callable or defines a callable!
• deletion is not yet a supported override

AI-assisted analysis of facebookresearch/detectron2@a2f4a8771a (2026-08-27). Data as JSON: /api/errors/097661428ecffa15. Report an issue: GitHub.