Document-parsing example

[WaterKnight1998]

Good morning,

First off all, thank you very much for open sourcing this model.

I have been looking at this model as an alternative to Donut for Document Parsing, I think we will get better performance as OCR data is included.

However, after checking your repository I just saw scripts for document classification and understanding. An example for Document Parsing or token classification will be helpfull. For document parsing I mean an example similar to this one for CORD dataset.

Thanks in advance! @zinengtang @ziyi-yang

Best regards

[ziyi-yang]

Thanks again for your interest in our work. Here's an example of how UDOP works on an actual document: https://github.com/microsoft/i-Code/blob/main/i-Code-Doc/example_io.ipynb

Thanks.

[thinh-huynh-re]

Hi @ziyi-yang , thank you for your amazing works. I follow the notebook with task_prefix = 'document classification', it outputs the final result as expected form However, when I change the task_prefix = 'layout analysis', it still outputs form So, Is the model only trained for one downstream task document classification ? (More information, I use the model udop-unimodel-large-224 downloaded from https://huggingface.co/ZinengTang/Udop)

[glahoti6]

I have a question same as @thinh-huynh-re. Could someone help us with that? Also, how to generate 00070353.json in the examples folder of i-Code-Doc?

[NielsRogge]

Hi folks, we made progress regarding this, it now works!

See https://discuss.huggingface.co/t/using-udop-for-layout-analysis/76871.