Improve Picture Extraction

[Leonerli2]

Hello everybody,

I really enjoy how i can use docling for my projects. I just have one problem with the picture extraction and wanted to know if anybody can help me here. I can extract some pictures out of PDFs just how I like it, but sadly it does not extract all of the pictures. So I wanted to know if there is a pipeline option, that can deal with that. I imagine a threshold option that I can set, so that it extracts an image, even if the model is not so certain.

My current pipeline option look like this:

Set up pipeline options for OCR and table structure

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = True
pipeline_options.images_scale = 4.0
pipeline_options.do_table_structure = True
pipeline_options.generate_picture_images = True # Enable picture extraction

I went through the pipeline option documentation, but I don't know what all the options could do. I really appreciate if anyone has a good idea how to tackle this.

Cheers!