What is Donut and why is OCR-free approach used?

Donut is a model for document processing that implements an OCR-free approach, meaning it does not use traditional optical character recognition (OCR) to extract text from document images.

Instead, Donut applies a Transformer architecture trained directly on document images and corresponding text annotations. This allows the model to understand the structure and content of the document as a whole, without the need to recognize individual characters first.

Advantages of the OCR-free approach:

• Reduction of errors related to character recognition
• Ability to work with different fonts, languages, and formats without additional OCR tuning
• Deeper understanding of the context and structure of the document

Thus, Donut combines visual perception and text generation in a single model, which improves the quality of information extraction from documents.