How to set CoordOrigin as TOPLEFT at DocumentConverter pipeline_options?

[JoeHelbing]

Curious if anyone know how to set CoordOrigin to be TOPLEFT instead of BOTTOMLEFT to pass to pipeline_options prior to DocumentConverter.

Every document I've tested with a .json export of the Docling Document object has produced a "BOTTOMLEFT" CoordOrigin, which I have to reconvert to "TOPLEFT" via a somewhat convoluted call to each prov element's BoundingBox.to_top_left_origin() function.

def flip_doc_y_axis(doc):
    # Create a deep copy to avoid modifying the original
    alt_doc = copy.deepcopy(doc)

    def convert_provs(item, page_heights):
        """Helper to convert all provenances for an item"""
        if hasattr(item, "prov"):
            for prov in item.prov:
                page_height = page_heights.get(prov.page_no)
                if page_height is not None:
                    prov.bbox = prov.bbox.to_top_left_origin(page_height) # BoundingBox.to_top_left_origin() function call

    # Get all page heights once
    page_heights = {
        page_no: page.size.height for page_no, page in alt_doc.pages.items()
    }

    # Convert coordinates in all document elements
    for text in alt_doc.texts:
        convert_provs(text, page_heights)

    for picture in alt_doc.pictures:
        convert_provs(picture, page_heights)

    for table in alt_doc.tables:
        convert_provs(table, page_heights)
        # Also convert table cell bounding boxes
        for cell in table.data.table_cells:
            if cell.bbox is not None:
                # Use the table's first provenance page height for cells
                if len(table.prov) > 0:
                    page_height = page_heights.get(table.prov[0].page_no)
                    if page_height is not None:
                        cell.bbox = cell.bbox.to_top_left_origin(page_height)

    for group in alt_doc.groups:
        convert_provs(group, page_heights)

    return alt_doc


# Runs the function to flip the document
flip_doc = flip_doc_y_axis(doc)

[cau-git]

@JoeHelbing DoclingDocument works by convention with BOTTOMLEFT origin coordinates as you found. However you can make the code to turn it into topleft coordinates much simpler by using the DoclingDocument.iterate_items method, you will only need one loop then.
Out of curiosity, for which use case do you need top-left coordinates in DoclingDocument?

[JoeHelbing]

For image cropping using PIL to match up with the boundingbox coordinates.

Made a gist of a guide to using docling for "geo-fenced" text extraction on standardized forms. I could stick with bottom left and translate between the two, but too much of a mind duck.

https://gist.github.com/JoeHelbing/17735d23f59efbfc894068a238d9f8df