Multilingual OCR Corpus Spanning 50 Languages

50.2 million image-text pairs across 50 languages support training and evaluation of multilingual text recognition systems. The dataset was created by Interfaze-AI and released on Hugging Face in late 2025. It is designed to handle diverse scripts and domains.

Use Cases

• Train multilingual text recognition models using image-text pairs across 50 languages.
• Evaluate OCR model performance on diverse scripts using the provided image and corresponding ground-truth text fields.
• Benchmark robustness of text detection systems against varied image domains and languages.

Strengths

• 50.2 million total image-text pairs.
• Covers 50 distinct languages.
• Designed for robust system evaluation across diverse scripts.

Limitations

• Specific row count per language and detailed column structure are unknown.
• Potential geographic or domain bias within the 50 languages is unspecified.
• Image quality, resolution, and annotation accuracy metrics are not provided.

Provenance

Source

Interfaze-AI via Hugging Face.

Collection Method

null

Time Range

null

Freshness

Dataset page was last updated on 2026-04-01.

Geography

null