Question about Table Extraction Model in Docling

[suyanecarvalho]

I’m currently trying to use the table extraction model from Docling in isolation and had a couple of questions:

Did the TableFormer model used by Docling undergo any fine-tuning? If so, could you provide details on the datasets or methods used for fine-tuning?

Could you explain how the table extraction model works in Docling? Specifically, how does it handle complex table structures, and what are the key components or steps involved in the process?

[dolfim-ibm]

You can find here the relevant publications of the TableFormer model:

• https://arxiv.org/abs/2203.01017
• https://arxiv.org/abs/2305.03393

[suyanecarvalho]

Hi Dolfin,Thank you for your response! I’ve read the articles and now have a good understanding of how the model works. However, I’m interested in using the table extraction model from Docling in isolation, without needing to download the entire Docling library.

Could you please explain if this is possible? If so, how can I achieve this? Are there specific components or packages I need to install, and are there any guidelines or examples for using the table extraction model standalone?