MMDoc: Multilingual Multimodal Document Translation Pairs

MMDoc is a dataset containing real-world document images with OCR results and translations across 10 language pairs. It is designed for training and evaluating multimodal document translation models that understand visual layout and context. The dataset was created by rileykim and was last updated on 2025-12-22.

Use Cases

• Train multimodal document translation models based on real-world document images and OCR results.
• Benchmark model performance on translating text in document images across 10 language pairs.
• Develop models that understand visual layout and context for document translation tasks.

Strengths

• Contains real-world document images, not synthetic data.
• Includes OCR results and translations for 10 language pairs.
• Designed specifically for multimodal translation tasks.

Limitations

• Row count, file formats, and column-level documentation are unknown.
• The dataset's license is unspecified, which may restrict usage.
• Freshness should be verified despite the 2025-12-22 update.

Provenance

Source

huggingface

Collection Method

Likely collected from real-world document sources, with OCR processing and translation applied.

Time Range

null

Freshness

Last updated 2025-12-22 16:15:38.

Geography

null