I’m unhappy with Google Vision OCR because it misses tables and key-value pairs

That’s a common pain point. Google Vision OCR is decent for plain text, but it’s not the best choice when you need document structure like:

• tables
• key-value pairs
• form fields
• reading order
• layout-aware extraction

Better options to consider

1. Google Document AI

If you want to stay in the Google ecosystem, this is usually the right upgrade.

• Better at layout analysis
• Has processors for forms, invoices, receipts, tables
• Can return structured output, not just raw OCR text

2. AWS Textract

Very strong for:

• tables
• forms / key-value pairs
• scanned documents

If your documents are business forms or invoices, Textract is often better than plain OCR.

3. Azure Document Intelligence (Form Recognizer)

Good for:

• structured document extraction
• forms
• tables
• custom models for your document types

4. Open-source alternatives

If you want control and lower cost:

• PaddleOCR + PP-Structure
• LayoutParser
• docTR
• Tesseract plus layout tools, though this is usually weaker for tables/forms

Practical recommendation

If your main issue is missing tables and key-value pairs, don’t use generic OCR alone. Use a document AI / layout extraction tool instead.

A good decision rule:

• Simple text images → OCR is fine
• Forms, invoices, tables, receipts → Document AI / Textract / Azure Document Intelligence
• Custom pipeline / self-hosted → PaddleOCR + layout analysis

If you want, I can help you with one of these:

1. compare Google Document AI vs Textract vs Azure
2. suggest the best option based on your document type
3. give you a sample pipeline for extracting tables and key-values

If you share what kind of documents you’re processing, I can recommend the best tool more precisely.