The lack of several functions is inconsistent with the paper.

[Idlerz7]

There is no description of the subsequent processing of images extracted from PDF files. There is no description of how image data is added to the knowledge graph. No code found for relation extraction.

[PascalSun]

currently in this release, the image is added to the KG via the layout KG.

SemanticKG Extraction from the Figures(Image) currently not included in this release.

We have a experiment version of Docs2KG, where the text of images will be extracted and go through similar SemanticKG extraction process like the text, it works fine for basic usage of retrieval, we are still evaluating whether this is ready for large scale adoption.

The direction for the image part we are currently working on is Image2Desc in json format with a model which is capable of understanding the figure and extract the information into a structured json format.

---

Semantic Relation Extraction is not included in this release, a general semantic relation extraction version will be released soon.

Hope that resolve your question.

Also, which domain you are working on and attempting to apply this, we are currently experimenting the Docs2KG on geology, financial, education and eye clinic, if you are working on a different domain, we can have further discussions regarding the problems to increase the generalibity of the KG construction process.