OCR dataset results

[cagey-squirrel]

Could you please share results from using OCR text extracting approaches on the datasets shared on HF? I am having trouble with PPOCR and would like to replicate results with text for RAG.

What would work best for me is the output of calling eval on ChartQA, InfoVQA, MP-DocVQA and SlideVQA datasets but with their text content being used for RAG instead of images (model used and OCR method does not matter too much, would prefer the one with best results).

[tcy6]

@cagey-squirrel

Could you please share results from using OCR text extracting approaches on the datasets shared on HF? I am having trouble with PPOCR and would like to replicate results with text for RAG.

I’m very sorry, but we couldn’t find the OCR results after conducting the experiments.

What would work best for me is the output of calling eval on ChartQA, InfoVQA, MP-DocVQA and SlideVQA datasets but with their text content being used for RAG instead of images (model used and OCR method does not matter too much, would prefer the one with best results).

Could you please explain this in more detail? I’m having trouble understanding it.

[cagey-squirrel]

Hi, thanks for the response.

When eval.sh script is called it produces embeddings.corpus, embeddings.query, test_result.log and test..rec files
It would be great if you had the test..trec files which are generated after running eval.sh on OCR versions of datasets ChartQA, InfoVQA, MP-DocVQA and SlideVQA.

[tcy6]

Hi, thanks for the response.

When eval.sh script is called it produces embeddings.corpus, embeddings.query, test_result.log and test..rec files It would be great if you had the test..trec files which are generated after running eval.sh on OCR versions of datasets ChartQA, InfoVQA, MP-DocVQA and SlideVQA.

Let me try to find it~
BTW, which OCR are you referring to?

[cagey-squirrel]

Not too important, if there are multiple then one with the best results.
Also I wanted to know how did you partition the text into chunks for RAG with OCR? Did you extract text page by page or did you group the chunks differently?

[tcy6]

Not too important, if there are multiple then one with the best results. Also I wanted to know how did you partition the text into chunks for RAG with OCR? Did you extract text page by page or did you group the chunks differently?

We extract text page by page.