Synthetic Korean Document Images for OCR-Free Understanding

Featuring 0.5 million synthetic Korean document images generated by the SynthDoG tool for training the Donut (OCR-Free Document Understanding Transformer) model. It was created by naver-clova-ix and last updated in January 2024.

Use Cases

• Train the Donut transformer model on synthetic Korean document images for OCR-free understanding tasks.
• Fine-tune vision-language models using the Korean text and layout features present in the SynthDoG-generated images.
• Benchmark document parsing and information extraction models on a large-scale synthetic Korean dataset.

Strengths

• Contains 0.5 million synthetic Korean document images, providing substantial scale for model training.
• Specifically designed for the Donut model, an OCR-free document understanding transformer published at ECCV 2022.
• Part of a multi-language suite including English, Chinese, and Japanese versions, each with 0.5 million samples.

Limitations

• Data is synthetically generated, which may not fully capture the noise and variability of real-world documents.
• The specific image characteristics, text categories, and annotation formats are not detailed in the provided input.
• The dataset's utility is tightly coupled with the Donut model's architecture and training objectives.

Provenance

Source

naver-clova-ix via Hugging Face.

Collection Method

Synthetically generated using the SynthDoG tool.

Freshness

Last updated on 2024-01-31.

Geography

Primarily Korean language content.