open-mmlab/mmdetection · warning

Zero area box skipped: {}.

Error message

Zero area box skipped: {}.

What it means

prefilter_boxes in WBF drops (continue) any box whose area (x2-x1)*(y2-y1) equals 0.0 and warns with the offending box_part. A zero-area box (a line or point, or a degenerate box where x1==x2 or y1==y2) cannot participate in IoU-based fusion, so it is excluded from the result. This is the only one of the three box checks that actually removes data.

Source

Thrown at mmdet/models/utils/wbf.py:171

            score = scores[t][j]
            if score < thr:
                continue
            label = int(labels[t][j])
            box_part = boxes[t][j]
            x1 = float(box_part[0])
            y1 = float(box_part[1])
            x2 = float(box_part[2])
            y2 = float(box_part[3])

            # Box data checks
            if x2 < x1:
                warnings.warn('X2 < X1 value in box. Swap them.')
                x1, x2 = x2, x1
            if y2 < y1:
                warnings.warn('Y2 < Y1 value in box. Swap them.')
                y1, y2 = y2, y1
            if (x2 - x1) * (y2 - y1) == 0.0:
                warnings.warn('Zero area box skipped: {}.'.format(box_part))
                continue

            # [label, score, weight, model index, x1, y1, x2, y2]
            b = [
                int(label),
                float(score) * weights[t], weights[t], t, x1, y1, x2, y2
            ]

            if label not in new_boxes:
                new_boxes[label] = []
            new_boxes[label].append(b)

    # Sort each list in dict by score and transform it to numpy array
    for k in new_boxes:
        current_boxes = np.array(new_boxes[k])
        new_boxes[k] = current_boxes[current_boxes[:, 1].argsort()[::-1]]

    return new_boxes

View on GitHub (pinned to cfd5d3a985)

Solutions

1. Filter degenerate boxes before fusion: keep = (b[:, 2] > b[:, 0]) & (b[:, 3] > b[:, 1]) then pass b[keep].
2. Clamp cxcywh sizes away from zero (w = max(w, eps), h = max(h, eps)) in your conversion code if zero-size boxes are spurious.
3. Inspect the emitting model if zero-area boxes appear frequently — it usually signals a broken bbox head, bad regression targets, or a data annotation problem.
4. If occasional drops are acceptable, ignore the warning: warnings.filterwarnings('ignore', message='Zero area box skipped.*').

Example fix

# before
labels, scores, boxes = weighted_boxes_fusion(boxes_list, labels_list, scores_list)
# after
boxes_list = [[b for b in bl if (b[2] - b[0]) > 0 and (b[3] - b[1]) > 0] for bl in boxes_list]
labels, scores, boxes = weighted_boxes_fusion(boxes_list, labels_list, scores_list)

def drop_degenerate(boxes, labels=None, scores=None):
    boxes = np.asarray(boxes)
    keep = (boxes[:, 2] - boxes[:, 0]) > 0
    if labels is None:
        return boxes[keep]
    return boxes[keep], [l for l, k in zip(labels, keep) if k], [s for s, k in zip(scores, keep) if k]

def all_positive_area(boxes):
    boxes = np.asarray(boxes)
    return bool(((boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1]) > 0).all())

When it happens

Trigger: Passing weighted_boxes_fusion() a boxes entry where x2 == x1 or y2 == y2-y1 == 0 — e.g. a clamped box at an image edge, a cxcywh box with w=0 or h=0, or float rounding that collapses a dimension to exactly 0.0. Note the exact == 0.0 comparison: only exactly-zero areas are skipped, not tiny ones.

Common situations: Models that regress zero width/height for very small or failed detections; boxes clipped to image bounds where the object is entirely outside so x1==x2 at the border; ensembling overconfidence-prone single-stage detectors that emit degenerate proposals.

Related errors

• X2 < X1 value in box. Swap them.
• Y2 < Y1 value in box. Swap them.
• LoadImageFromFile is not found in the test pipeline
• Visualization needs the "visualizer" termdefined in the conf
• Unsupported input type: {type(single_input)}

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/1af6a6bbdeef8e59. Report an issue: GitHub.