Synthetic Arabic Text Images for OCR Training

SARD is a large-scale synthetic dataset for Arabic Optical Character Recognition. It provides controlled and diverse training data simulating real-world book layouts. The dataset was created by riotu-lab and was last updated in April 2026.

Use Cases

• Train OCR models to recognize Arabic characters from synthetic image features.
• Evaluate model performance on diverse, controlled Arabic text layouts.
• Benchmark OCR accuracy using the dataset's simulated book layout features.
• Develop layout analysis algorithms using the dataset's structured page features.

Strengths

• Dataset is described as large-scale and synthetically generated.
• Data is designed to be diverse and scalable for training.

Limitations

• All quantitative details like row count, image count, and size are unknown.
• Data is synthetic, which may not fully capture real-world noise and variations.

Provenance

Source

riotu-lab

Collection Method

Synthetically generated.

Freshness

Last updated in April 2026.