Tatar OCR Benchmark with Page Images and Layout Annotations

Tatar-language documents form a page-level benchmark for optical character recognition. Each item includes the original page image, structured OCR and layout annotations in JSONL, and a visual control image for verification. The dataset was created by yasalma and last updated on March 31, 2026.

Use Cases

• Benchmarking OCR model performance on Tatar text based on the provided page images and annotations.
• Training layout analysis models using the structured JSONL annotations for document regions.
• Verifying OCR output quality by comparing the original and reconstructed visual control images.
• Developing multimodal document understanding systems for a low-resource language.

Strengths

• Includes full original page images for direct model input.
• Provides structured OCR and layout annotations in JSONL format.
• Contains visual control images for fast human verification of results.

Limitations

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

Provenance

Source

yasalma on Hugging Face, with acknowledgements to Yandex LLC for support in collection.

Collection Method

Likely collected and annotated from Tatar documents; specific method is not detailed.

Freshness

Last updated 2026-03-31 18:36:22; freshness should be verified.