Synthetic Arabic Document Pages with HTML Ground Truth for OCR Training

A large-scale, synthetically generated dataset for training and evaluating document understanding and Optical Character Recognition models. The dataset was created by FatimahEmadEldin and last updated on HuggingFace in September 2025. It provides page images paired with their corresponding structured HTML ground truth, with a focus on the Arabic language.

Use Cases

• Train Optical Character Recognition models based on synthetic Arabic page images.
• Evaluate document layout understanding models based on structured HTML ground truth.
• Benchmark multilingual OCR performance based on the focus on the Arabic language.
• Develop data augmentation pipelines based on the synthetically generated nature of the dataset.

Strengths

• Dataset is described as large-scale, suggesting substantial volume for training.
• Focus on Arabic language provides a resource for a specific script.
• Data pairs images with structured HTML ground truth, providing dual-modality supervision.

Limitations

• Row count, file formats, and column definitions are unknown, limiting suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.
• The synthetic generation method may not fully capture the noise and variation of real-world documents.

Provenance

Source

HuggingFace, created by FatimahEmadEldin.

Collection Method

Synthetically generated.

Time Range

null

Freshness

Last updated 2025-09-21 04:41:52; freshness should be verified.

Geography

null