Exception when process some image detect 0 height tables and possible fix

[bash99]

First of all, I would like to express my gratitude for the two outstanding open-source projects, marker and surya_ocr.

When I process a specific single image, for example, by running the following command:

marker_single --output_format markdown --disable_image_extraction data/out/bad_9.jpg

I encounter the following error:

  File "...lib/python3.11/site-packages/surya/input/processing.py", line 43, in slice_polys_from_image
    lines.append(slice_and_pad_poly(image_array, poly))
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "...lib/python3.11/site-packages/surya/input/processing.py", line 58, in slice_and_pad_poly
    cv2.fillPoly(mask, [np.int32(coordinates)], 1)
cv2.error: OpenCV(4.11.0) :-1: error: (-5:Bad argument) in function 'fillPoly'
> Overload resolution failed:
>  - Layout of the output array img is incompatible with cv::Mat
>  - Expected Ptr<cv::UMat> for argument 'img'

I attempted to debug the parameters passed to fillPoly, and the relevant context is as follows:

    # Pad the area outside the polygon with the pad value
    mask = np.zeros(cropped_polygon.shape[:2], dtype=np.uint8)
    print(coordinates)
    print(cropped_polygon.shape[:2])
    cv2.fillPoly(mask, [np.int32(coordinates)], 1)

The output is:

coordinates = [(0, 0), (367, 0), (367, 78), (0, 78)]
cropped_polygon.shape[:2] = (0, 367)

It appears that the height of the mask is 0, which caused the fillPoly function to throw an error.

To address this, I modified the code as follows:

    # Pad the area outside the polygon with the pad value
    mask_shape = cropped_polygon.shape[:2]
    # If mask shape is 0, it means the polygon is outside the image, so we return the full image
    if(mask_shape[0] == 0 or mask_shape[1] == 0):
        return Image.fromarray(image_array)
    mask = np.zeros(cropped_polygon.shape[:2], dtype=np.uint8)

This modification avoids the error while returning a basically correct result. However, since I am not very familiar with the overall structure of marker, I am unsure whether this fix is reasonable. If the fix is acceptable, could it be merged into the upstream? If needed, I can also create a pull request.

Unfortunately, I am unable to upload the image that triggers the error because I spent dozens of minutes trying to redact private information from it. However, once the private information is removed, the exception no longer occurs. If the image is absolutely necessary, I may need to obtain authorization and then consider sending it via email.

Thank you for your understanding and support!