Arabic Text Images for OCR, Over 2 Million Labeled Samples

Over 2.16 million labeled images of Arabic text extracted from diverse sources, created by mssqpi and last updated on March 14, 2025. The dataset is designed to enhance Optical Character Recognition capabilities for the Arabic language and is stored in Parquet format with a total size of 1.87 GB.

Use Cases

• Training Arabic OCR models based on the large collection of labeled text images.
• Benchmarking OCR model performance on diverse Arabic text sources mentioned in the description.
• Developing multilingual OCR systems that include Arabic language support.
• Fine-tuning pre-trained vision-language models for Arabic text understanding.

Strengths

• Contains over 2.16 million labeled samples, providing substantial training data.
• Total file size of 1.87 GB indicates a significant volume of image-text pairs.
• Images are extracted from diverse sources, which may improve model generalization.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.
• The specific sources and potential biases in the text extraction are not detailed.

Provenance

Source

huggingface user mssqpi

Collection Method

Extracted from diverse sources (unspecified).

Freshness

Last updated 2025-03-14 15:44:02