Nom Ocr Data: Historical Hán-Nôm Manuscript Pages for OCR Training

An ongoing corpus of historical Hán-Nôm manuscript pages annotated for OCR training. Each page includes a high-resolution image, per-character bounding boxes with corrected text labels, and candidate alternates from two upstream OCR engines. The dataset is maintained by Aerbote88 and was last updated on May 11, 2026.

Use Cases

• Train OCR models for historical scripts based on per-character bounding boxes and corrected text labels.
• Compare OCR engine performance using candidate alternates from Kandianguji and Nôm Na Việt engines.
• Analyze document structure based on reading-order column polygons for text, binding, marginalia, and commentary.
• Study character-level uncertainty and metadata for improving OCR confidence scoring.

Strengths

• Includes per-character bounding boxes with corrected text labels, providing precise ground truth.
• Contains candidate alternates from two distinct upstream OCR engines for comparison.
• Features reading-order column polygons that segment text, binding, marginalia, and commentary.
• An ongoing project updated frequently with new data, as of May 2026.

Limitations

• Row count, file formats, and dataset size are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• License information is unknown, which may restrict usage.

Provenance

Source

Historical Hán-Nôm manuscripts.

Collection Method

Annotated pages with per-character bounding boxes and text labels; includes outputs from Kandianguji and Nôm Na Việt OCR engines.

Freshness

Last updated 2026-05-11 00:04:57; described as an ongoing project updated frequently.