domains / families / documents / openocr

openocr

separate

Extract text from images, documents and scanned PDFs using OpenOCR

When to use

• Extract text from images
• Extract text from documents
• Extract text from scanned PDFs
• Text detection
• Text recognition
• Universal VLM recognition
• Document parsing with layout analysis
• Recognize formulas and tables

When not to use

• Text recognition accuracy depends on image quality
• Very small or heavily rotated text may reduce accuracy
• `server` mode requires PyTorch and is slower than `mobile` mode
• UniRec and Doc tasks use 0.1B parameter VLM, larger models may yield better results
• PDF processing converts pages to images internally, very large PDFs may use significant memory
• Complex handwritten text accuracy varies
• GPU recommended for best performance, especially for Doc and UniRec tasks

Skill page

Source