Published August 5, 2026 | Version v1
Dataset Open

SAQR: A Synthetic-to-Real Arabic Handwriting Recognition, Retrieval, and Demographic Dataset

Authors/Creators

  • 1. ROR icon Chungbuk National University
  • 2. ROR icon Assiut University
  • 3. ROR icon University of Louisville

Description

SAQR is a benchmark dataset for Arabic Handwritten Text Recognition (HTR), Printed-to-Handwritten Retrieval/Matching, and Demographic (Gender) Analysis. 

The dataset contains 2,263 aligned pairs of synthetic printed ground-truth crops and their handwritten counterpart line images written by diverse writers across 3 official splits (Train: 1,575, Validation: 335, Test: 353).

Files

SAQR_Dataset.zip

Files (291.9 MB)

Name Size
md5:a8838bc9ca1e68596dee13755c6634a6
291.9 MB Preview Download