NayanaOCR Corpus 2025: 1M-Page Multilingual Synthetic Document Dataset

1 million pages of fully-parallel synthetic documents rendered in 22 languages for OCR, layout detection, and visual question answering tasks. The dataset was created by Cognitive-Lab and was last updated on the Hugging Face platform in May 2026. It is described as one of the largest open-source multilingual, multi-task document datasets, with the same ~45,700 source pages rendered in every language.

Use Cases

• Training optical character recognition (OCR) models based on the 1 million synthetic document pages.
• Evaluating layout detection algorithms using the parallel document structure across 22 languages.
• Developing visual question answering (VQA) systems for documents based on the described multi-task corpus.
• Benchmarking multilingual document understanding models on low-resource languages mentioned in the description.

Strengths

• Large scale with 1 million document pages.
• True parallel corpus structure across 22 languages.
• Designed for multiple tasks: OCR, layout detection, and VQA.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.
• Data is synthetic, which may not fully capture the characteristics of real-world documents.

Provenance

Source

Cognitive-Lab on Hugging Face.

Collection Method

Synthetic generation; the description states it is a 'synthetic OCR + VQA corpus'.

Freshness

Last updated 2026-05-25 16:45:43; freshness should be verified.