The possibility of 100% local deployment

[bit-scientist]

I am quite overwhelmed by the amount of tasks this repository supports, but still unable to wrap my head around this.

I would like to know if it is possible to build a pipeline that is developed (deployed) 100% locally.
I have the following scenario:

1. I have around 500k scanned document jpg images that I need to do an OCR. Almost all of these images contain scanned doc texts (typed and handwritten) and tables (official forms, checks, etc.) of varying quality, and they are in Korean. See a sample image
2. Similarly, I have 200k digitally generated PDF (pure PDF) that are mostly selectable/searchable text data and tables of various forms.

My job is to extract data from both formats (jpg and pdf) as accurately as possible in order for me to build a DB that can be later referenced or queried. I am more concerned about the structural preservation of table data in both jpg and pdf files, because we will build a RAG-based Q&A app in the future.

Thanks.

[liuhongen1234567]

Hello, based on your description, you can use the PP-Structurev3 pipeline to obtain structured data. If you are more focused on tables, you can use the Table Pipeline v2.
The reference document link is as follows:
PP-StructureV3： https://www.paddleocr.ai/main/en/version3.x/pipeline_usage/PP-StructureV3.html
General Table Recognition v2： https://www.paddleocr.ai/main/en/version3.x/pipeline_usage/table_recognition_v2.html