Skip to content

Convoluted Reading Flow for Multi-Column PDFs #1203

@LuRe97

Description

@LuRe97

Bug

When giving Docling PDFs with mainly two column layouts as an input, the reading order gets messed up for multiple instances. Examples are Headings or paragraphs that get attributed to the reading flow of the other column, images being attributed as child of another heading or even paragraphs of the same text column being switched up in the markdown result. At the worst it even mixed up the reading order of one heading element ("Key Features" ---docling-parse----> "Features Key")

Example Files:
Data Sheet Example 1
Data Sheet Example 2

Steps to reproduce

Load the files via the basic example:

from docling.document_converter import DocumentConverter

source =  # document URL
converter = DocumentConverter()
result = converter.convert(source)
print(result.document.export_to_markdown())

Docling version

Docling version: 2.27.0
Docling Core version: 2.23.2
Docling IBM Models version: 3.4.0
Docling Parse version: 4.0.0
Python: cpython-311 (3.11.10)
Platform: Linux-5.15.0-94-generic-x86_64-with-glibc2.35

Python version

Python 3.11.10

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions