New Sinhala OCR Dataset and Model Achieve State-of-the-Art Performance

Researchers have developed a new dataset, sinhala-ocr-lk-acts-1010, to improve Optical Character Recognition (OCR) for the Sinhala language, which is spoken by approximately 16 million people in Sri Lanka. This dataset comprises 1,010 page-level images and their transcriptions, sourced from Sri Lankan Legislative Acts spanning two decades. Experiments fine-tuning three deep learning models, including DeepSeek-OCR V1, DeepSeek-OCR V2, and LightOnOCR-2-1B, demonstrated that LightOnOCR-2-1B achieved the best performance with a character error rate of 1.05%, significantly outperforming other open-source and commercial OCR solutions. AI

IMPACT Improves OCR capabilities for low-resource languages, potentially aiding historical document digitization and accessibility.

RANK_REASON Academic paper introducing a new dataset and OCR model evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• DeepSeek-OCR V1
• DeepSeek-OCR V2
• Google Document AI
• LightOnOCR-2-1B
• Nevidu Jayatilleke
• QLoRA
• Sinhala
• sinhala-ocr-lk-acts-1010
• Sri Lanka
• Surya-OCR
• Tesseract v5

• paper
• other

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