New VLM extracts document data without OCR, outperforming larger models

Researchers have developed a new method for extracting key-value pairs from document images without relying on traditional OCR preprocessing. They fine-tuned a compact 256M-parameter vision-language model called SmolDocling to perform this task end-to-end, jointly handling identification, localization, and association. This approach reportedly outperforms larger zero-shot VLMs on benchmarks like FUNSD and XFUND, while being significantly smaller and faster than models like Qwen2.5-VL. AI

IMPACT This approach could streamline document processing pipelines and improve efficiency in information extraction tasks.

RANK_REASON The cluster contains an academic paper detailing a new model and methodology for document image analysis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• FUNSD
• Hugging Face
• Qwen2.5-VL
• SmolDocling
• X

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →