Researchers have introduced Persian Pixel, a large-scale synthetic dataset designed to improve Optical Character Recognition (OCR) for the Persian language. The dataset contains over 343,000 image-text pairs, generated using the SynthOCR-Gen framework to accurately model the complexities of the Persian script, including cursive connectivity and stylistic variations. Persian Pixel aims to overcome the scarcity of annotated data, enabling the training of advanced OCR models like TrOCR and Donut and advancing Persian document analysis. AI
IMPACT This dataset could significantly improve AI's ability to process and understand Persian text, aiding in digitization and analysis of historical documents.
RANK_REASON The item is a research paper introducing a new dataset for a specific language's OCR. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
