Benchmark on OmniDocBench

It will be cool to compare your pipeline against others on [OmniDocBench](https://github.com/opendatalab/OmniDocBench). You can run with gemini model, I guess it will be okay.


Actually, I manage [pdf-extraction-agenda](https://github.com/dantetemplar/pdf-extraction-agenda/), and I have added your pipeline to it.