ValueError: "EasyOcrOptions" object has no field "coverage_threshold"

[AdityaMannu1709]

Can we have a sample code snippet for the coverage threshold part in the documentation?

pipeline_options = PdfPipelineOptions()
pipeline_options.images_scale = IMAGE_RESOLUTION_SCALE
pipeline_options.generate_page_images = False
pipeline_options.generate_table_images = False
pipeline_options.generate_picture_images = True
pipeline_options.ocr_options.coverage_threshold = 0.05

doc_converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
    }
)

[dolfim-ibm]

Do you mean explaining the parameter?

• This parameters controls the minimum area ratio (with respect to the page) for a picture to be processed with OCR.
• In short: if (image_area / page_area) > pipeline_options.ocr_options.coverage_threshold: DO OCR ON IT

The title of your post "object has no field coverage_threshold" looks a bit suspicious? Are you having problem with this parameter?

• Since it was recently added, I would suggest making sure you are using the latest docling version.
• The actual parameter is called pipeline_options.ocr_options.bitmap_area_threshold.

[AdityaMannu1709]

Can we be provided with a sample snippet for ignoring smaller images like logos bcz using this I m unable to skip those?

[sanjana-7890]

Yes I also have the same query....... it would be great if anyone shares the example code snippet for skipping certain below threshold in PDFs so that along with image skipping those skipped image's placeholder is also skipped in doctag or markdown.

[dolfim-ibm]

It should be something like this

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = True
# Choose the threshold that works best for you. Default is 0.05
pipeline_options.ocr_options.bitmap_area_threshold = 0.2

doc_converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
    }
)

result = doc_converter.convert(...)

# To remove (all) the image placeholders
result.document.export_to_markdown(image_placeholder="")

[AdityaMannu1709]

Yes that is working but only in markdown output schema ......... ideally it should be applicable on doctags also.
The <figure></figure> tags should also be removed just like image placeholders are being removed in markdown considering threshold.

Shouldn't it be like that in doctags bcz they pose advantage over markdown when u r dealing with captions coverage for images so removing irrelevant images in doctags is also necessary?

[dolfim-ibm]

You could try something like this

from docling_core.types.doc import DocItemLabel
from docling_core.types.doc.document import DEFAULT_EXPORT_LABELS

result.document.export_to_document_tokens(labels=DEFAULT_EXPORT_LABELS - {DocItemLabel.PICTURE})