opendatalab/MinerU · error · ValueError

Shape of table bounding boxes is not between in 4 or 8.

Error message

Shape of table bounding boxes is not between in 4 or 8.

What it means

Raised by the UNet table recognizer's visualization routine when table_cell_bboxes has a second dimension that is neither 4 (axis-aligned boxes: x1,y1,x2,y2) nor 8 (quad polygons: 4 xy points). The drawing code dispatches on bboxes.shape[1] and can only draw rectangles (4) or polylines (8).

Source

Thrown at mineru/model/table/rec/unet_table/utils.py:381

    ):
        if save_html_path:
            html_with_border = self.insert_border_style(table_results.pred_html)
            self.save_html(save_html_path, html_with_border)

        table_cell_bboxes = table_results.cell_bboxes
        table_logic_points = table_results.logic_points
        if table_cell_bboxes is None:
            return None

        img = self.load_img(img_path)

        dims_bboxes = table_cell_bboxes.shape[1]
        if dims_bboxes == 4:
            drawed_img = self.draw_rectangle(img, table_cell_bboxes)
        elif dims_bboxes == 8:
            drawed_img = self.draw_polylines(img, table_cell_bboxes)
        else:
            raise ValueError("Shape of table bounding boxes is not between in 4 or 8.")

        if save_drawed_path:
            self.save_img(save_drawed_path, drawed_img)

        if save_logic_path:
            polygons = [[box[0], box[1], box[4], box[5]] for box in table_cell_bboxes]
            self.plot_rec_box_with_logic_info(
                img, save_logic_path, table_logic_points, polygons
            )
        return drawed_img

    def insert_border_style(self, table_html_str: str):
        style_res = """<meta charset="UTF-8"><style>
        table {
            border-collapse: collapse;
            width: 100%;
        }
        th, td {

View on GitHub (pinned to 4fe4bde114)

Solutions

1. Ensure the array passed for cell bbox drawing is np.ndarray with shape (N, 4) or (N, 8).
2. If your detector emits (N, 5) with scores, slice them off: bboxes = arr[:, :4].
3. Check upstream post-processing (e.g. output bbox decoder) hasn't been replaced or reordered.

Example fix

# before
drawed = table_rec.visualize(img_path, raw_pred[:, :5])  # score column kept

# after
bboxes = raw_pred[:, :4] if raw_pred.shape[1] > 4 else raw_pred
drawed = table_rec.visualize(img_path, bboxes)

import numpy as np
bboxes = np.asarray(table_cell_bboxes)
if bboxes.ndim != 2 or bboxes.shape[1] not in (4, 8):
    raise ValueError(f"expected (N,4) or (N,8) bboxes, got {bboxes.shape}")

def is_valid_bboxes(arr) -> bool:
    arr = np.asarray(arr)
    return arr.ndim == 2 and arr.shape[1] in (4, 8)

When it happens

Trigger: Calling the draw/save-visualization API of the table model with predicted cell boxes shaped (N, 5) (e.g. boxes with confidence appended), (N, 6), or feeding raw model output without post-processing to box format; also (0,) shaped arrays from an empty prediction reshaped incorrectly.

Common situations: Custom post-processing that appends a score column to each bbox; passing logic_points (4-point logical indices) where pixel bboxes were expected; passing a transposed or flattened array.

Related errors

• Invalid scale {scale}, must be positive.
• Input image ({w}, {h}) smaller than the target size ({cw}, {
• {model_path} does not exists.
• {model_path} is not a file.
• Input must be a pillow object or a numpy array.

AI-assisted analysis of opendatalab/MinerU@4fe4bde114 (2026-08-14). Data as JSON: /api/errors/fcf392dae207d6b3. Report an issue: GitHub.