roboflow/supervision · error · ValueError

LabelMe polygon shape (label={label}) has {len(points)} poin

Error message

LabelMe polygon shape (label={label}) has {len(points)} point(s); expected at least 3.

What it means

Raised by the LabelMe dataset loader when a non-rectangle shape (polygon, linestrip-like types treated as polygons) contains fewer than 3 points. A polygon needs at least 3 vertices to enclose an area and be converted to a bounding box, so the loader refuses to convert it. The offending shape's label is included in the message to help locate it in the JSON annotation file.

Source

Thrown at src/supervision/dataset/formats/labelme.py:127

            )
        points = np.array(points_raw, dtype=np.float32)
        if points.ndim != 2 or points.shape[1] != 2:
            raise ValueError(
                f"LabelMe shape of type {shape_type!r} (label={label!r}) has "
                f"malformed points: expected an (N, 2) array, got shape "
                f"{points.shape}."
            )
        if shape_type == "rectangle":
            if len(points) < 2:
                raise ValueError(
                    f"LabelMe rectangle shape (label={label!r}) has "
                    f"{len(points)} point(s); expected at least 2."
                )
            xyxy = _rectangle_to_xyxy(points)
            polygon = _xyxy_to_polygon(xyxy)
        else:
            if len(points) < 3:
                raise ValueError(
                    f"LabelMe polygon shape (label={label!r}) has "
                    f"{len(points)} point(s); expected at least 3."
                )
            xyxy = polygon_to_xyxy(polygon=points).astype(np.float32)
            polygon = points
        xyxy_list.append(xyxy)
        class_ids.append(class_to_index[label])
        if with_masks:
            polygons.append(polygon)

    if skipped_types:
        warnings.warn(
            f"Skipped unsupported LabelMe shape type(s) {sorted(skipped_types)}; "
            f"only {list(SUPPORTED_SHAPE_TYPES)} are imported.",
            UserWarning,
            stacklevel=2,
        )

View on GitHub (pinned to 7f254d9784)

Solutions

Open the LabelMe JSON file named in the surrounding error context and find the shape with the reported label; add a third point or delete the shape.
If the shape was meant to be a line, either convert it to shape_type 'rectangle' with 2 diagonal points or filter it out before loading.
Write a small pre-processing pass over the dataset that drops or repairs shapes with len(points) < 3 before calling the loader.
Regenerate the annotation from the source tool (e.g. re-draw in LabelMe) if the file is corrupted.

Example fix

// before (labelme JSON)
{"label": "door", "shape_type": "polygon", "points": [[10, 10], [40, 10]]}

// after
{"label": "door", "shape_type": "polygon", "points": [[10, 10], [40, 10], [40, 40]]}

Defensive patterns

Strategy: validation

Validate before calling

import json

def valid_labelme_shapes(path):
    with open(path) as f:
        data = json.load(f)
    for shape in data.get("shapes", []):
        pts = shape.get("points", [])
        if shape.get("shape_type") == "rectangle":
            if len(pts) < 2:
                return False
        elif len(pts) < 3:
            return False
    return True

Try / catch

try:
    ds = sv.DetectionDataset.from_labelme(...)
except ValueError as e:
    if "expected at least 3" in str(e):
        log.warning("skipping malformed annotation: %s", e)
    else:
        raise

Prevention

Run a dataset lint pass that flags shapes with fewer points than their shape_type requires before loading.
Prefer re-exporting from the LabelMe tool rather than hand-editing points arrays.
In CI for annotation datasets, assert every shape has >= 3 points (>= 2 for rectangles).

When it happens

Trigger: Loading a LabelMe dataset (DetectionDataset.from_labelme or the labelme format module) where a shapes[] entry with shape_type other than 'rectangle' has a points array of length 0, 1, or 2. Happens with hand-drawn annotations, corrupted files, or exports from tools that emit degenerate polygons.

Common situations: Manually annotated datasets where a shape was started but not finished; third-party converters that emit empty points arrays; JSON edits that truncated points; linestrip shapes (2 points) being treated as polygons.

Related errors

AI-assisted analysis of roboflow/supervision@7f254d9784 (2026-08-15). Data as JSON: /api/errors/13be58e12d78df03. Report an issue: GitHub.