- Structure Evaluation Mannequin: A mannequin primarily based on RT-DETR and educated on DocLayNet (a human-annotated knowledge set for doc structure evaluation) that classifies web page parts like paragraphs, part titles, lists, and tables.
- TableFormer: A vision-transformer mannequin for desk construction restoration that may deal with complicated tables with partial or no borderlines, empty cells, cell spans, and hierarchical headers.
The Docling processing pipeline works by feeding web page pictures to the Structure Evaluation Mannequin, which identifies doc parts. For tables, TableFormer processes the detected desk areas to get well their construction. When wanted, OCR capabilities can be found via integration with EasyOCR.
Utilizing Docling is simple:
from docling.document_converter import DocumentConverter
supply = "https://arxiv.org/pdf/2408.09869" # doc per native path or URL
converter = DocumentConverter()
consequence = converter.convert(supply)
print(consequence.doc.export_to_markdown()) # output: "## Docling Technical Report[...]"
Docling additionally supplies a handy command-line interface for fast conversions:
docling https://arxiv.org/pdf/2206.01062
Key use instances for Docling
Docling’s capabilities make it very best for a number of important use instances together with retrieval-augmented technology, data base creation, LLM fine-tuning, and enterprise knowledge integration.
