PDFtoTextConvertor ocr option not working

[jais001]

Hi,

I am working on PDF to text conversion. But while using OCR argument in PDFToTextConverter, I am getting an error, i.e

unexpected keyword argument 'ocr'

I have checked the api documentation of PDFToTextConverter and found this argument in the documentation. My haystack version is

farm-haystack ==1.17.0

Does the argument exist in the latest version or is this an error?

[anakin87]

Hello @JaisVJ!

I am on Google Colab and the following commands work for me.
For more information, please take a look at the installation guide.

! pip install farm-haystack[pdf,ocr]==1.17.0    # You should select the right dependencies groups.
! apt-get install tesseract-ocr    # Tesseract is needed for OCR

# I set the environment variable that points at Tesseract data
# The following command works in Ubuntu. For other operating systems, you should use different commands.
import os
os.environ["TESSDATA_PREFIX"]="/usr/share/tesseract-ocr/4.00/tessdata"

converter = PDFToTextConverter(valid_languages=['en'], ocr='auto')

[jlncoelho]

Should haystack really be quietly failing to import the correct class and instead importing a different one with the same name?

try:
    with LazyImport() as fitz_import:
        # Try to use PyMuPDF, if not available fall back to xpdf
        from haystack.nodes.file_converter.pdf import PDFToTextConverter  # type: ignore

    fitz_import.check()
except (ModuleNotFoundError, ImportError):
    from haystack.nodes.file_converter.pdf_xpdf import PDFToTextConverter  # type: ignore  # pylint: disable=reimported,ungrouped-imports

[anakin87]

Hello...
I understand your concern but you have 2 different options for PDF conversion, as explained in the docs. They have different license.

In Haystack 2.0 (now in beta), there are different and more explicit options.