Images of documents get extracted as document text

[Bain-OS]

Bug

Figures/images that contain text are incorrectly exploded as content, rather than just an image.

Steps to reproduce

For example, here is some debug output attained by executing docling --debug-visualize-cells --debug-visualize-layout https://arxiv.org/pdf/2408.09869

Notice the figure is an image of a document. Docling is detecting the text within the image as text, I would have expected just two elements on this page, the image and its description.

Is there a means to achieve this with configuration?

Docling version

> docling --version
Docling version: 2.15.1
Docling Core version: 2.14.0
Docling IBM Models version: 3.1.2
Docling Parse version: 3.0.0

Python version

Python 3.11.10

[wcool1]

https://github.com/opendatalab/MinerU