Is there any way i can stop table parsing about pdf2docx

[dothinking]

I'm afraid no, at least for now. Table has two roles in this library: lattice table for structure data, and stream table in case float layout (e.g. multi-paragraphs in same row). So, I guess the "false positive tables" are stream tables, which are required by reproducing the layout.

Could you please share a test case? Maybe we can find a workaround or to make some change accordingly.

[monuminu]

thanks for replying . I think better option would be have a flag to identify steam table only , lattice table only , stream lattice both , none. In this way we can definitely have much more control about document conversion . unfortunately I can't share the document as it's organisation internal document .

[dothinking]

May I know your purpose using this lib, the pdf text only, or both text and layout (e.g. paragraph spacing, indentation and text format)? If you concern only the text, I'd suggest to use PyMuPDF directly, or similar pdf processing libs like pikepdf, PyPDF, pdfminer...

The lib extracts pdf text with an upstream lib PyMuPDF, and focuses on layout rebuilding, while stream table is responsible for the layout (split float layout into flow layout in each cell). The lib thinks it's necessary to if stream table exists in your docx. In this case, if turn it off, the layout rebuilding may fail.

[dothinking]

close this for now. feel free to reopen it if any new thoughts.

[monuminu]

Hey is there any way i can only have lattice table detection and switch off stream table detection .

[dothinking]

Hey is there any way i can only have lattice table detection and switch off stream table detection .

As described before, stream table is required to rebuild the page layout. But if you don't care much about the layout, yes, you can switch it off. For now, you need to do it manually -> hard comment stream table parsing. Take version 0.5.1 for example,

1. go to file Layout.py

>>> import pdf2docx
>>> pdf2docx.page.Layout.__file__
'd:\\21_github\\pdf2docx\\pdf2docx\\page\\Layout.py'

2. go to method _parse_layout_bottom_up() (Line 119), then hard-comment stream table parsing as needed.

def _parse_layout_bottom_up(self, settings:dict):
    '''Parse layout bottom-up: 
    
    * detect explicit tables first based on shapes, 
    * then stream tables based on original text blocks and parsed explicit tables;
    * move table contained blocks (text block or explicit table) to associated cell layout.
    '''
    # parse table structure/format recognized from explicit shapes
    self._table_parser.lattice_tables(
                    settings['connected_border_tolerance'],
                    settings['min_border_clearance'],
                    settings['max_border_width'])
    
    # parse table structure based on implicit layout of text blocks
    self._table_parser.stream_tables(
                    settings['min_border_clearance'],
                    settings['max_border_width'],
                    settings['float_layout_tolerance'],
                    settings['line_separate_threshold'])

I think better option would be have a flag to identify steam table only , lattice table only , stream lattice both , none. In this way we can definitely have much more control about document conversion.

I concerned much about the function of stream table for page layout. But as suggested, flexible option of table parsing does accommodate more use cases. As seen in method _parse_layout_bottom_up, it's easy to add such flag in settings. Would do it in next version.

Thank you so much for this good suggestion.

[dothinking]

v0.5.2 is available in Pypi now, arguments parse_lattice_table and parse_stream_table, both True by default, should work for this issue. The usage might look like:

from pdf2docx import Converter

pdf_file = '/path/to/sample.pdf'
docx_file = 'path/to/sample.docx'

# parse lattice tables only
cv = Converter(pdf_file)
cv.convert(docx_file, parse_stream_table=False)
cv.close()