Kuzushiji Dataset: Historical Japanese Documents with Character Annotations

Japanese historical documents (Kotenseki) containing page images with character-level bounding box annotations for Kuzushiji (cursive Japanese) recognition. The dataset was created by Kotomiya07 and was last updated on March 8, 2026.

Use Cases

• Train optical character recognition (OCR) models for Kuzushiji script based on character-level bounding boxes.
• Develop document layout analysis tools for historical Japanese texts using page images.
• Benchmark machine learning models for historical document understanding tasks.
• Study the evolution of Japanese cursive writing styles across different historical periods.

Strengths

• Includes character-level bounding box annotations, which are precise for training recognition models.
• Dataset structure includes image IDs and book IDs, suggesting organization by source document.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Historical Japanese documents (Kotenseki).

Freshness

Last updated 2026-03-08 21:34:56; freshness should be verified.

Geography

Japan