Docatlas Instruct: Multilingual OCR Dataset Across 82 Languages

DocAtlas provides high-fidelity multilingual OCR data covering 82 languages and 10 writing systems. The dataset is built through model-free differential rendering and provides precise structural annotations in a unified DocTag format. It was created by ahmedheakl and last updated in May 2026.

Use Cases

• Train multilingual OCR models based on the dataset's coverage of 82 languages.
• Develop document layout analysis systems based on the unified structural annotations.
• Benchmark document understanding models across diverse writing systems.
• Fine-tune models for specific document types based on the annotated component types.

Strengths

• Covers 82 languages and 10 writing systems, enabling broad multilingual evaluation.
• Provides precise structural annotations in a unified DocTag format without relying on learned models.
• Built through model-free differential rendering, which likely ensures high-fidelity ground truth.

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

huggingface

Collection Method

Built through model-free differential rendering.

Freshness

Last updated 2026-05-20 00:40:49; freshness should be verified.